Photodegradation of ciprofloxacin by tannin-barium titanate catalysts and evaluation via artificial neural network modeling approaches

Abstract

The visible-light-driven removal of ciprofloxacin (CIP) remains a challenge due to its persistence and contribution to antibiotic resistance. In this study, tannin-modified barium titanate (BaTiO₃) composites were synthesized using three different tannin forms (formaldehyde tannin, calcined tannin, and calcined formaldehyde tannin) to enhance photocatalytic performance. UV-DRS analysis confirmed that tannin modification effectively reduced the band gap of BaTiO₃ from ~ 3.4 eV to 3.15 eV (FTB) and ~ 2.6 eV (C-TB and C-FTB), enabling efficient visible-light activity. Under optimal operating conditions (pH 3, 5 ppm CIP, 20 mg catalyst), formaldehyde tannin-modified BaTiO₃ (FTB) achieved a maximum degradation efficiency of ~ 95% within 180 min. Kinetic evaluation showed that CIP degradation followed pseudo-first-order kinetics (k_app = 0.0086 min⁻¹ for FTB), and isotherm modelling indicated the best fit to the Freundlich model, confirming multilayer adsorption on heterogeneous surfaces. The artificial neural network (ANN) model demonstrated high predictive accuracy (R² > 0.99), identifying catalyst type and initial CIP concentration as the most influential parameters. These results highlight tannin-modified BaTiO₃ composites as promising, low-cost, and eco-friendly visible-light photocatalysts for antibiotic removal. The innovative aspect of this study is the modification of BaTiO₃ with bio-derived tannin derivatives. Tannin addition offers a sustainable and low-cost approach while also accelerating CIP degradation by reducing electron recombination. Additionally, the experimental data were integrated with an advanced ANN model, providing a powerful and interpretable hybrid approach for predicting catalyst performance.

Supplementary Information

The online version contains supplementary material available at 10.1038/s41598-025-34946-9.

Introduction

Antibiotics are drugs used to treat infections caused by bacteria. Antibiotics are consumed on the world market at a scale of 100,000-200,000 tons every year¹. Fluoroquinolones are one of the largest and most frequently prescribed classes of antibiotics and are largely discarded intact after use². Ciprofloxacin (CIP) is a widely used fluoroquinolone antibiotic and is therefore frequently detected in wastewater and it is very important to remove them from the aquatic ecosystems where they are found. These antibiotics that enter the aquatic ecosystem cause antibiotic resistance in both the aquatic organisms and humans. Therefore, these and similar antibiotics need to be removed from wastewater^1–4. The accumulation of pharmaceutical residues poses severe risks to aquatic ecosystems and public health, necessitating the development of advanced treatment technologies for effective antibiotic removal^3,4. Antibiotics are difficult to remove from wastewater with traditional methods due to their low biodegradability and cause accumulation in aquatic environments. Therefore, alternatively, high-efficiency, easy-to-apply, environmentally friendly and economical methods need to be developed and methods such as adsorption, photocatalytic degradation, biodegradation and chemical oxidation come to the fore⁵.

In order to remove CIP from water environments, more efficient methods are sought other than traditional methods. In recent studies, photocatalytic degradation stands out among the methods used for this purpose. The basic mechanism of photocatalytic processes is the breakdown of the targeted waste by hydroxyl (·OH) radicals to form CO₂ and inorganic ions. When the electrons on the photocatalyst surface are exposed to a photon higher than the band gap energy of the photon energy, they change places and move from the valence band to the conduction band. After photon absorption, positively charged holes and electrons moving towards the conduction band are formed on the photocatalyst surface. Redox reactions occur between the electrons and the positively charged holes. The electrons react with water or oxygen molecules adsorbed on the surface of the photocatalyst and form ·OH radicals or superoxide radicals () with high oxidation power⁶.

In recent years, photocatalysis has attracted great attention due to its high reaction rates, free radical reactions and non-toxicity. The most difficult part of this process is the recovery of the catalyst. It is known that this problem can be overcome by doping or immobilizing the photocatalytic materials with different materials. Photocatalytic processes are considered a promising option for the treatment of wastewater containing pollutants such as CIP⁷.

Barium titanium oxide (BaTiO₃) is a material with potential for use as a photocatalyst in the removal of different pollutants in aqueous environments. It has photocatalytic activity because it is a semiconductor material. Although it is generally effective under UV light, it is known to exhibit high performance under visible light when modified with different materials. In addition, BaTiO₃ has the chemical stability expected from catalysts in photocatalytic reactions. It is considered an advantageous option for applications in the removal of pollutants from wastewater due to its environmentally friendly material and other advantages listed. In cases where it is used in pure form, its photocatalytic performance is limited due to the difficulties encountered in the recovery and reuse of the catalyst and the need for UV light excitation due to the wide band gap mentioned earlier. Modification methods such as doping or composite formation should be applied to increase photocatalytic activity^8–10.

Tannins are phenolic compounds that can be naturally found in plants. It is thought that they can be considered as an important option for increasing photocatalytic performance. The reason for this can be said to be that they facilitate electron transfer due to the phenolic and aromatic groups in their structures, stabilize electron gaps and narrow the band gap. For this purpose, there are studies where they are used in modifying photocatalysts such as BaTiO₃, TiO₂ and ZnO. In these studies, it has been reported that the addition of tannin has a positive effect on photodegradation efficiency. At the same time, being a natural biological material, not being toxic and being naturally degradable in nature allows the photocatalytic system they are used in to be economically and environmentally sustainable^11–14. Recent studies have further demonstrated the potential of novel biomass waste-based photocatalysts for the efficient degradation of organic pollutants, highlighting the growing importance of eco-friendly synthesis routes¹⁵. Artificial Neural Network (ANN) is one of the computational intelligence models in machine learning and has become an important tool especially in the field of analyzing such processes.

This method is inspired by the structure of human brain and processing of information by biological neural networks, which contributes to the modeling of the relationship between input variables and system output, allowing the understanding of systems that are difficult to solve analytically. Conventional statistical experimental design methods have been widely used for modelling and optimizing chemical processes. For instance, Response Surface Methodology (RSM) has been successfully employed to optimize effective factors in photocatalyst fabrication and antibiotic removal¹⁶. However, these methods could have limited efficiency regarding to highly nonlinear processes¹⁷. Therefore, machine learning based approaches have gained importance since their ability to process information from complex systems. Recent literature demonstrates that harnessing artificial intelligence and advanced optimization algorithms significantly enhances the prediction accuracy and process efficiency for the degradation of organic pollutants^18–20. To this day, literature studies revealed ANN is successfully used in optimization studies of chemical processes such as pollutant removal and system optimization. ANNs are trained using weights and biases through interconnected layers and neurons in these layers and are optimized to minimize the prediction lines during this process. In this way, they can successfully predict the removal processes of various pollutants from wastewater with high accuracy and contribute to the optimization of these processes^21–23.

In this study, the removal of CIP from aqueous media by photocatalytic degradation under visible light was investigated using BaTiO₃ composites modified with different tannin sources (formaldehyde tannin, calcined tannin, and calcined formaldehyde tannin). The primary novelty of this work lies in the integration of this natural polymer-assisted photocatalyst modification with advanced data-driven modeling. Specifically, the incorporation of formaldehyde-tannin introduces covalent crosslinking that stabilizes the tannin structure and promotes electron transfer across the BaTiO₃ interface, resulting in enhanced visible-light activity and reduced recombination losses compared to traditional methods. Along with the synthesis and characterization, the effects of various experimental parameters (initial concentration, pH, time, catalyst dosage) were examined, followed by detailed isotherm and kinetic studies. Complementing the experimental work, an ANN model was developed not only to predict removal efficiency but to offer a robust evaluation using multiple performance metrics (R², MSE, MAE, SEP, MRPD). This approach distinguishes the present study from prior single-metric modeling efforts by ensuring higher reliability and interpretability. Ultimately, this hybrid experimental–computational framework provides both mechanistic understanding and predictive capacity, revealing the comprehensive removal performances of these new eco-friendly catalysts.

Materials and methods

Materials

Barium chloride dihydrate (BaCl₂·2 H₂O, ≥ 99% purity, Merck) and analytical-grade ethanol (≥ 99.8%, Merck) were used in the synthesis. Titanium(IV) butoxide (Ti(OBu)₄, 97% purity, Merck) served as the titanium precursor without further purification. The commercial tannin extract used for BaTiO₃ modification was supplied by Artu Kimya (Manisa, Turkey). According to the supplier specifications, the extract consisted of approximately 70–75% tannin, along with moisture (8–10%), inorganic ash (2–5%), and non-tannin organic matter (remainder). The extract was utilized directly in the synthesis process without any additional treatment.

Characterization

The detailed characterization of the samples was conducted using advanced analytical techniques to ensure comprehensive understanding of their properties. Fourier Transform Infrared Spectroscopy (FTIR) was performed with a Bruker VERTEX 70v, enabling the identification of functional groups and chemical bonds present in the materials. X-Ray Diffraction (XRD) analysis, carried out using a PANalytical Empyrean XRD, provided critical information on the crystalline structure and phase composition. Surface area and porosity measurements were obtained through Brunauer–Emmett–Teller (BET) analysis using a Micromeritics 3Flex, allowing for the assessment of material adsorption capabilities. Furthermore, the optical properties were investigated using UV-Visible Diffuse Reflectance Spectroscopy (UV-DRS) with a Shimadzu UV-3600 Plus, offering insights into the optical behavior and band gap energy. These state-of-the-art techniques collectively delivered valuable and precise information on the structural, morphological, and optical properties of the samples, facilitating a deeper understanding of their potential applications.

Synthesis of photocatalysts

BaTiO₃ composites obtained with different tannin additives were prepared to be used in the removal of CIP from aqueous media. In this study, tannin-doped BaTiO₃ catalyst was synthesized (FTB) containing 10% tannin or formaldehyde tannin by weight based on the weight of barium. In order to examine the effect of the calcination process on this process, these catalysts were subjected to the calcination process. Thus, calcined tannin BaTiO₃ (C-TB) and formaldehyde tannin BaTiO₃ (C-FTB) catalysts were synthesized. Tannin and BaTiO₃ catalyst (TB) were not used in this study because it had been examined in a previous study¹⁴. First of all, in order to prepare formaldehyde tannin doped catalysts, formaldehyde tannin must be synthesized from tannin. 50 ml of ammonia (13.3 N) solution was added to the tannin and mixed. Then, 65 ml of formaldehyde solution (37%) was added to the ammonia tannin solution. Centrifugation was performed to separate the solid formed. 100 ml of nitric acid (0.1 N) was added to the resulting precipitate and then dried in the oven. Thus, formaldehyde tannin was obtained²⁴.

In the preparation of tannin- or formaldehyde-tannin-modified BaTiO₃ photocatalysts, a sol–gel-based synthesis route was followed, in line with methodologies reported in earlier studies¹⁴. Initially, 20 mL of titanium butoxide was combined with 40 mL of ethanol to form the titanium precursor solution. Separately, barium chloride dihydrate was dissolved in distilled water. These two solutions were then mixed together. In another step, tannin or formaldehyde-tannin added in an amount equal to 10% of the barium content was dissolved in 10 mL of water and incorporated into the mixture. The resulting solution underwent hydrothermal treatment in an autoclave at 200 °C for 24 h. After cooling, the solid product was collected, thoroughly rinsed with deionized water, and dried at 80 °C for one day. These steps yielded TB and FTB photocatalysts, while additional heat treatment at 700 °C for 24 h was applied to produce the calcined forms C-TB and C-FTB.

Photocatalytic experiments

The ability of the synthesized catalysts to degrade ciprofloxacin was assessed under illumination in the visible spectrum. For this purpose, light sources emitting in the 400–800 nm range were employed. Prior to initiating the photocatalytic reaction, all solutions were kept in the dark for a set duration to establish adsorption–desorption equilibrium. Once a stable state was reached, the light sources were activated, marking the beginning of the photocatalytic process.

The degradation behavior of ciprofloxacin was explored under continuously mixed conditions in a laboratory setting. Throughout the reaction period, aliquots were collected at intervals and passed through a filter to remove suspended solids. The optical properties of these clear samples were examined by recording their absorbance values at 277 nm using a UV spectrophotometer²⁵. Based on a previously established calibration curve, these absorbance readings were transformed into concentration values, allowing for the calculation of degradation efficiency through the applied formula Eq. (1)⁷.

1

Within the employed equation, C_i indicates the starting concentration of ciprofloxacin, while C_t represents the concentration at a given reaction time. The photocatalytic behaviors of BaTiO₃-based catalysts, synthesized through various tannin-derived modifications, were systematically evaluated under visible light. Moreover, the influence of operational factors-such as catalyst dosage, pH, and reaction time-was analyzed to better understand their contributions to the overall degradation efficiency.

ANN modelling studies

In order to predict the removal performance of ciprofloxacin via photodegradation by barium titanate modified tannin composites, Artificial Neural Network modelling studies were performed by neural network toolbox (nntool) in MATLAB version R2109a. The influence of five different input parameters including photocatalyst type, pH, initial concentration of CIP, contact time and photocatalyst dosage on removal efficiency as output parameter was assessed. A three-layer network configuration with 5 input neurons and 1 output neuron was optimized by varying number of neurons at hidden layer within the range of 1 and 20. Activation function for both hidden and output layers were chosen as tangent sigmoid transfer function (tansig) and Levenberg-Marquardt backpropagation (Trainlm) with 1000 iterations was used for training of network (Table S1 and Table S2). Data samples (N = 117) were randomly grouped into training (70%), validation (15%), and testing (15%) sets using MATLAB’s default random sampling procedure to avoid bias. The model design and evaluation followed the standard procedure of supervised learning, where the training set was used for weight adjustment, the validation set for preventing overfitting, and the test set for performance assessment. Performance of the trained network was evaluated considering the mean square error (MSE) and regression coefficient (R²) values expressed by Eq. (2) and Eq. (3) respectively, where t_i is the targeted value, p_i is the predicted value, N is the number observations and t_m is the average of targeted values²³. Additionally, mean absolute error (MAE) (Eq. 4), standard error of prediction (SEP) (Eq. 5), mean relative percentage deviation (MRPD) (Eq. 6) and adjusted R² (R²_adj) (Eq. 7) metrics were calculated for further commenting on the model performance, where RMSE stands for root mean square error and p for the number of independent variables²⁶. All used input and target data were normalized in the range of -1 and 1 according to Eq. (8), where x is the variable value, x_norm is the normalized value, x_min and x_max are the minimum and maximum variable values respectively²⁰. maximum [-1, 1] interval has been selected since the ANN model employed tansig activation function, which operates effectively in the case of inputs centering around zero. The Details of the selected ANN modelling parameters were summarized in Table 1.

Table 1.

Details of selected ANN modelling parameters.

Parameters	Definitions
Training algorithm	Feed-forward backpropagation (Trainlm)
Activation function of hidden layer	tansig
Activation function of output layer	tansig
Performance function	MSE, MAE, SEP, MRPD
Number of input neurons	5
Number of output neurons	1
Number of iterations	1000
Learning rate	0.01

2

3

4

5

6

7

8

In order to predict the singular influence of each input parameters on ciprofloxacin removal, sensitivity analysis was carried out by the algorithm proposed by Garson^21,22. The weight values obtained from the ANN model calculations were used for the evaluation of parameters given in Eq. (9), Eq. (10) and Eq. (11), where C is the contribution of each input neuron and R is the relative contribution of each input neuron.

9

10

11

Isotherm studies for CIP photodegradation

In order to provide insights into CIP removal by modified composites, adsorption isotherm models were adapted for the fitting of experimental photodegradation data. Modelling was conducted according to three different models namely Langmuir, Freundlich and Temkin isotherms, which allowed to better understand the nature of CIP photodegradation by tannin/BaTiO₃ composites. Relevancy to Langmuir model indicates monolayer adsorption on a homogeneous surface with the assumption of noninteracting adsorbate particles^27,28. The nonlinear and linearized demonstrations of Langmuir model are given in Eq. (12) and Eq. (13) respectively, where q_m represents the maximum Langmuir rate constant which can be associated to estimated maximum photodegradation capacity²⁹. The C_e (mg/L) and K_L (L/mg) stand for equilibrium concentration of CIP and Langmuir isotherm constant respectively.

12

13

The K_L constant of Langmuir model and C₀ value ((mg/L) the maximum utilized initial CIP concentration) is used to calculate the R_L factor depicted in Eq. (14), which could be applied for assessing the favorability of adsorption process, hence also the favorability of photodegradation. A R_L factor of 0 is evidence of irreversible, 0 < R_L<1 is favorable and R_L>1 is unfavorable adsorption process³⁰.

14

Freundlich isotherm model is associated with multilayer adsorption on a heterogeneous surface and presented by Eq. (15):

15

where K_F (mg/g) is the Freundlich model constant related to adsorption capacity and n to adsorption intensity. n value in the range of 1 and 10 is accepted as an indicator of good adsorption process and could be used for assessing the favorability of photodegradation when isotherm is applied to photodegradation data²⁷. Linearized expression of the Freundlich model Eq. (16) is followed for model creation by plotting ln(q_e) versus ln (C_e).

16

The nonlinear Temkin isotherm model is represented by the following Eq. (17):

17

where K_T (L/g) is the Temkin isotherm constant and b (kJ/mol) is related to heat of degradation²⁹. Linearized expression of Temkin isotherm is given in Eq. (18).

18

Kinetic studies for CIP photodegradation

In this study, the kinetics of the removal process of ciprofloxacin by photodegradation was investigated for all three synthesized composites. Since it is known that the Langmuir-Hinshelwood (L-H) model is widely used in the kinetic investigation of such processes, this model was also used in this study^31,32. The Eq. (19) for the L-H model:

19

When the concentration of the studied solution is high, Eq. (19) reduces to a zero-order kinetic model (Eq. 20) since the K_C term is much greater than 1.

20

However, at low initial concentrations, the L–H model (Eq. 19) follows a pseudo-first-order kinetic behavior, as expressed in Eq. (21).

21

In these equations; C₀: initial ciprofloxacin concentration, C: concentration at any moment of degradation, k: rate constant of the process, k_app: apparent rate constant, K: adsorption equilibrium coefficient, t: exposure time to illumination.

Results & discussion

FTIR analysis

As we examine the FTIR characterization of the catalysts, distinct differences in functional groups are evident among FTB, C-TB, and C-FTB, as shown in Fig. 1. Firstly, the broad peak around 3200 cm⁻¹, attributed to hydroxyl (OH) groups, is prominent in the FTB catalyst but noticeably absent in both C-TB and C-FTB. This difference suggests that the calcination process reduces the presence of hydroxyl groups, as reported in studies where heating removes surface hydroxyls, which can otherwise act as adsorption sites and impact the catalyst’s surface properties³³. Additionally, a strong peak at approximately 1600 cm⁻¹ is observed in the FTB sample, indicating the presence of carbonyl (C = O) stretching vibrations. Carbonyl groups are typically associated with organic compounds or intermediate species that can remain unaltered in non-calcined samples. The absence of this peak in the calcined C-TB and C-FTB samples aligns with prior findings showing that calcination can decompose organic residues, leading to a cleaner surface free from such functional groups³⁴​. Furthermore, both C-TB and C-FTB exhibit a peak around 2150 cm⁻¹, which is absent in FTB. This peak may correspond to alkyne (C ≡ C) groups or other triple-bonded carbon structures that could form during the calcination process. The formation of these groups may indicate structural changes in the catalyst’s carbon framework under high-temperature treatment, potentially enhancing specific catalytic properties by introducing new reactive sites³⁵​. The calcination process significantly influences the structural and surface characteristics of catalysts, which in turn affects their photocatalytic performance. Specifically, the reduction of hydroxyl and carbonyl groups in calcined samples like C-TB and C-FTB can alter their interaction with pollutants, potentially enhancing stability under reaction conditions. Increasing calcination temperatures led to a decrease in surface hydroxyl groups, which correlated with changes in photocatalytic activity³⁶.

Fig. 1.

FTIR spectra showing functional group differences among FTB, C-TB, and C-FTB.

XRD analysis

The XRD analysis of the catalysts reveals significant differences in their structural characteristics. As shown in Fig. 2, the XRD pattern of FTB, broad and low-intensity peaks indicate a predominantly amorphous and disordered structure, which is typical of organic compounds with main elements likely being carbon, hydrogen, and oxygen. This lack of crystallinity suggests that FTB does not exhibit the organized lattice structure seen in inorganic materials³⁷. The XRD patterns of C-TB and C-FTB display sharp and distinct peaks, signifying a high degree of crystallinity. Notable diffraction peaks are observed at approximately 25°, 32°, 47°, and 56°, corresponding to well-crystallized phases consistent with barium titanate (BaTiO₃), an inorganic compound known for its stable crystal structure. These findings suggest that the main elements present in C-TB and C-FTB are barium, titanium, and oxygen, which contribute to their structured, crystalline nature. This observation aligns with previous studies on barium titanate, where similar XRD patterns have been reported, confirming the material’s crystalline structure³⁸. Overall, this comparison highlights the difference in structural composition, with FTB being primarily amorphous, while C-TB and C-FTB exhibit the crystalline characteristics of barium titanate. This crystallinity in C-TB and C-FTB may enhance their stability and catalytic performance due to the ordered arrangement of atoms in the crystal lattice³⁹.

Fig. 2.

XRD patterns: For the catalysts FTB, C-TB, C-FTB and pure BaTiO₃.

XPS analysis

The chemical purity and elemental composition of the synthesized photocatalysts were initially assessed using XPS survey scans, as presented in Fig. 3. The wide-scan spectra for FTB, C-TB, and C-FTB clearly display the characteristic photoelectron peaks corresponding to the constituent elements: Barium (Ba 3d, Ba 4d), Titanium (Ti 2p), Oxygen (O 1s), and Carbon (C 1s).

Fig. 3.

XPS survey scan spectra of the synthesized composites: (a) FTB, (b) C-TB, and (c) C-FTB, indicating the elemental composition of Ba, Ti, O, and C.

As observed in the spectra, the strong peaks located at binding energies of roughly 780 eV (Ba 3d), 458 eV (Ti 2p), and 530 eV (O 1s) confirm the preservation of the main BaTiO₃ host structure across all modification methods (tannin incorporation and calcination). Furthermore, the distinct C 1s peak detected at ~ 285 eV in all samples evidences the successful incorporation of the organic tannin modifier into the composite structure. No significant peaks related to other impurities were detected, confirming the high purity of the synthesized materials.

X-ray Photoelectron Spectroscopy (XPS) was conducted to investigate the surface elemental composition and chemical states of the FTB, C-TB, and C-FTB composites. The high-resolution spectra for Ba 3d, C 1s, O 1s, and Ti 2p core levels are presented in Fig. 4.

Fig. 4.

High-resolution XPS spectra of the synthesized photocatalysts: (a–d) FTB composite, (e–h) C-TB composite, and (i–l) C-FTB composite. The spectra in each row correspond to Ba 3d, C 1s, O 1s, and Ti 2p core levels, respectively.

Figure (4a, e & i) shows the Ba 3d spectra for all samples. Two distinct peaks appear at binding energies of approximately 779.5 eV and 794.9 eV, corresponding to the spin–orbit splitting of Ba 3d₅/₂ and Ba 3d₃/₂, respectively. These values confirm the presence of Ba²⁺ ions within the perovskite lattice structure of BaTiO₃⁴⁰. The sharp and symmetrical peaks observed in the calcined samples (C-TB and C-FTB) indicate a well-defined crystalline environment, consistent with the XRD results.

The Ti 2p spectra (Fig. 4d, h & l) exhibit two characteristic peaks centered around 458.4 eV (Ti 2p₃/₂) and 464.2 eV (Ti 2p₁/₂). The spin–orbit splitting energy of approximately 5.8 eV confirms that titanium exists in the Ti⁴⁺ oxidation state in all composites⁴¹. No significant shift in peak positions was observed between samples, suggesting that the tannin modification did not notably alter the core electronic environment of the titanium ions.

The O 1s spectra (Fig. 4c, g & k) can be deconvoluted into two main contributions. The dominant peak at lower binding energy (~ 529.6 eV) corresponds to lattice oxygen (O_L) in the Ba–O and Ti–O bonds of the BaTiO₃ crystal lattice. The shoulder at higher binding energy (~ 531.5–532 eV) is attributed to surface-adsorbed oxygen species (O_S), such as hydroxyl groups (–OH), oxygen vacancies, or oxygen bonded to carbon (C–O/C = O) originating from the tannin structure⁴². Notably, these surface oxygen species are crucial for photocatalysis, as they facilitate pollutant and water molecule adsorption and promote the formation of reactive radicals.

The C 1s spectra (Fig. 4b, f & j) display a major peak around 284.8 eV, associated with adventitious carbon and C–C/C = C bonds in the tannin skeleton. The asymmetry and shoulders at higher binding energies (~ 286–289 eV) indicate the presence of oxygen-containing carbon functionalities, such as C–O (hydroxyl/ether) and C = O (carbonyl) groups⁴³. In the FTB sample (Fig. 4b), these functionalities confirm the successful incorporation of the formaldehyde–tannin polymer network onto the BaTiO₃ surface without thermal degradation. Even in the calcined samples (C-TB and C-FTB), the persistence of carbon signals suggests that a carbonaceous layer or doped carbon species remains, which may enhance conductivity and visible-light absorption.

UV-DRS analysis

The band gap energies of the catalysts were investigated using UV-DRS analysis to assess their potential photocatalytic performance. The typical band gap range for semiconductors lies between 2.0 eV and 3.2 eV. Titanium dioxide (TiO₂), one of the most widely used photocatalysts, has a band gap of 3.2 eV⁴⁴, while barium titanate (BaTiO₃) has a slightly higher band gap of approximately 3.4 eV⁴⁵. Through doping with tannin and formaldehyde tannin, the band gap of the catalysts was effectively reduced, allowing them to harness a broader spectrum of visible light. This reduction in band gap facilitates the generation of more electron-hole pairs, thereby enhancing the production of reactive species that contribute to improved photocatalytic degradation of ciprofloxacin⁴⁶. As shown in Fig. 5b, the UV-DRS analysis of C-TB revealed a band gap of approximately 2.6 eV, as determined by extrapolating the linear portion of the Tauc plot to the x-axis. This band gap characterizes C-TB as a wide-band-gap semiconductor, which is well-suited for applications that require the absorption of high-energy photons. FTB, on the other hand, exhibits a band gap of around 3.15 eV (Fig. 5a). Although this band gap is relatively high, FTB maintains robust photocatalytic activity due to its ability to generate a substantial number of electron-hole pairs under visible light irradiation. Additionally, FTB’s favorable surface properties, including a large surface area and an abundance of active sites, contribute to its effectiveness in degrading ciprofloxacin⁴⁷. The band gap of C-FTB was also found to be approximately 2.6 eV, (Fig. 5c) which is within the optimal range for visible light absorption. However, despite its comparable band gap, C-FTB’s photocatalytic efficiency is somewhat limited by its smaller surface area and fewer active sites relative to FTB. While both catalysts are effective in the degradation of ciprofloxacin, the difference in surface area and active site density between FTB and C-FTB results in a performance variation, with FTB achieving higher degradation efficiency due to its superior surface characteristics⁴⁸.

Fig. 5.

UV-DRS analysis of FTB (a), C-TB (b), and C-FTB (c).

PL analysis

As shown in Fig. 6, the photoluminescence (PL) intensities of FTB, C-TB, and C-FTB exhibit clear differences that directly reflect their charge-carrier behaviors. The markedly lower PL emission intensity observed for the C-FTB sample indicates a significant suppression of electron–hole recombination, suggesting that the formaldehyde–tannin modification effectively enhances charge separation and prolongs carrier lifetime. In contrast, the higher PL intensity of FTB implies faster recombination, consistent with its relatively lower photocatalytic efficiency. These results align with recent studies showing that reduced PL intensity is strongly correlated with more efficient charge transport and improved visible-light photocatalytic performance^49,50. Overall, the PL profiles confirm that the C-FTB catalyst possesses superior charge-carrier management, which supports its higher CIP degradation activity.

Fig. 6.

PL emission spectra of (a) FTB, (b) C-TB, and (c) C-FTB composites.

BET analysis

The Brunauer–Emmett–Teller (BET) analysis, as illustrated in Fig. 7 for the catalysts (a) FTB, (b) C-TB, and (c) C-FTB, provides insight into the surface area and porosity of the synthesized catalysts, which are critical factors influencing their photocatalytic performance. FTB exhibits a significantly higher surface area of 174.74 m²/g and a pore volume of 0.29 cm³/g, suggesting a substantial number of active sites available for catalytic reactions. In contrast, C-TB and C-FTB have lower surface areas of 21.93 m²/g and 27.43 m²/g, respectively, indicating fewer active sites.

Fig. 7.

BET isotherm plots for FTB (a), C-TB (b), and C-FTB (c).

Regarding pore size distribution, C-TB and C-FTB possess larger average pore sizes of 273.57 nm and 212.45 nm, respectively, classifying them as mesoporous materials. Conversely, FTB has a smaller average pore size of 34.23 nm, indicating a microporous structure. These structural differences suggest that FTB’s higher surface area and microporous nature may enhance its photocatalytic efficiency by providing more active sites and facilitating better adsorption of reactant molecules. This observation that higher surface area and pore volumes are associated with an increase in surface active sites and channels, leads to improved photocatalytic activity⁵¹.

SEM analysis

The surface morphologies of the FTB, C-TB, and C-FTB composites were examined using SEM to understand the effects of formaldehyde incorporation and calcination on the microstructure. Distinct morphological features were observed across the samples, largely consistent with the BET results presented in Fig. 8. The FTB composite synthesized without calcination exhibited an amorphous and spongy structure composed of irregularly shaped, loosely packed particles. Numerous interconnected pores and fine voids were visible throughout the surface, indicating the presence of a highly porous and open network. This morphology indicates that the polymeric formaldehyde-tannin skeleton was largely preserved. The SEM findings are consistent with the strikingly high surface area and small average pore volume obtained from the BET analysis, confirming that the absence of calcination favors the formation of fine, uniformly distributed pores that enhance surface accessibility and reactivity. In contrast, the C-TB composite exhibited a compact morphology with large, densely packed crystalline domains typical of BaTiO₃. The particles exhibited smoother surfaces and limited visible porosity, indicating collapse of the organic phase during calcination. The well-defined angular shapes observed in the micrographs reflect high crystallinity. The large average pore size measured by BET supports this observation and indicates that only a few macropores remained after the degradation of the tannin-derived matrix during calcination. The C-FTB composite exhibited an intermediate morphology between FTB and C-TB. The surfaces were rougher and more irregular, with smaller fragments and fine particles scattered across the BaTiO₃ grains. The addition of formaldehyde during resin formation appears to have improved the dispersion of the carbonaceous phase and partially alleviated the pore collapse caused by calcination. Accordingly, the results observed in the BET analysis correlate well with the SEM images. Overall, the SEM observations clearly agree with the BET results. The uncalcined FTB composite retained a highly porous and open texture with small and evenly distributed pores, while the calcined samples became denser with progressively fewer and larger pores. Among these, C-FTB retained greater structural roughness and smaller pores due to the stabilizing effect of formaldehyde. These results confirm that both the calcination process and formaldehyde incorporation play important roles in determining the pore structure and surface properties of BaTiO₃-based composites.

Fig. 8.

SEM images of (a) FTB, (b) C-TB and (c) C-FTB composites.

Zeta potential

Zeta potential measurements were performed at pH 3–7 to determine the surface charge properties and isoelectric points (IEPs) of the composites, and the results are presented in Fig. 9. In all samples, the zeta potential decreased as pH increased, and the surface charge shifted from positive to negative. This trend is known to be related to the balance of protonation and deprotonation of the surface hydroxyl groups. In one study, BaTiO₃ particles exhibited a positive zeta potential at pH 2–3. However, as pH increased, the potential decreased, crossing the zero point around pH 4–5, and remained negative thereafter until pH 8. Another study on tannin-based compounds indicated that the surface properties exhibited negative zeta potential over a wide pH range after pH 4–5, depending on the modification method. However, in modifications using formaldehyde and ammonium salts, the addition of amine and quaternary ammonium groups to the tannin chains has been reported to increase the surface’s positive charge density, shifting the isoelectric point to higher pH values. The zeta potential results for the composites obtained in this study are consistent with the literature. The FTB sample initially exhibited a positive surface charge, but with increasing pH, the potential gradually decreased to negative values. The C-TB sample initially exhibited a higher positive charge, but similar to FTB, it approached zero around pH 4–5, reaching the isoelectric point in this region. The surface charge of the C-FTB composite remained positive between pH 3–5, with the charge approaching zero at pH 6. This difference is thought to be due to the combined application of both formaldehyde and calcination modifications. Consequently, zeta potential measurements revealed a trend toward increasing positive surface charge in the order FTB < C-TB < C-FTB.

Fig. 9.

Zeta potential values of FTB, C-TB and C-FTB composites measured at different pH values.

Effect of initial Ciprofloxacin concentrations

The degradation efficiency of ciprofloxacin using C-TB, FTB, and C-FTB catalysts shows a clear dependence on the initial concentration of the pollutant. As shown in Fig. 10a, at 5 ppm, all three catalysts exhibit high performance, with C-FTB and FTB slightly outperforming C-TB, indicating their higher reactivity under low contaminant conditions. At 10 ppm, C-FTB becomes the most efficient catalyst, followed by FTB and C-TB (Fig. 10b). At moderate CIP concentrations (10–20 ppm), the C-FTB sample exhibits comparatively higher degradation efficiency, which is attributed to its modified surface chemistry facilitating improved charge separation and pollutant adsorption. However, similar to the other catalysts, its performance decreases at concentrations above 20 ppm due to light-screening effects and active-site saturation. At 20 ppm, the performance of C-TB decreases significantly, whereas FTB and C-FTB maintain their high degradation rates (Fig. 10c). This trend continues at 40 ppm, where FTB shows the highest efficiency, followed by C-FTB, while C-TB exhibits the lowest degradation efficiency (Fig. 10d). These findings suggest that FTB and C-FTB are more robust and efficient at managing higher pollutant concentrations, making them better suited for applications involving higher or fluctuating contaminant levels, as compared to C-TB⁴⁸.

Fig. 10.

Effect of Initial Ciprofloxacin Concentrations on Degradation Efficiency: (a) 5 ppm, (b) 10 ppm, (c) 20 ppm, (d) 40 ppm, with Catalysts C-TB, FTB, and C-FTB.

Effect of different catalyst amounts

When evaluating the effect of catalyst amounts on the degradation efficiency of ciprofloxacin using C-TB, FTB, and C-FTB, a clear trend emerges. As shown in Fig. 11a, varying the catalyst amounts from 5 mg to 40 mg reveals distinct performances for each catalyst. At a lower catalyst amount (5 mg), FTB demonstrates the highest degradation efficiency, surpassing both C-FTB and C-TB. At a lower catalyst amount (5 mg), FTB demonstrates the highest degradation efficiency, surpassing both C-FTB and C-TB. This superior performance at low dosage is attributed to the exceptionally high specific surface area of FTB (174.74 m²/g) compared to C-TB (21.93 m²/g) and C-FTB (27.43 m²/g), as confirmed by BET analysis. The highly porous and spongy morphology of FTB ensures that even a small mass of catalyst provides abundant accessible active sites for the pollutant molecules. In contrast, the calcined samples (C-TB and C-FTB), having significantly lower surface areas, require higher dosages to provide sufficient active surface sites for effective degradation. However, as the catalyst amount increases to 20 mg and 40 mg, C-FTB becomes the most efficient catalyst (Fig. 11b). At a lower catalyst amount (5 mg), FTB demonstrates the highest degradation efficiency, attributed to its exceptionally high specific surface area (174.74 m²/g). However, its performance plateaus as the catalyst amount increases (Fig. 11c), suggesting that the available active sites are already sufficient even at low dosages (Xu et al., 2023). On the other hand, C-FTB benefits significantly from increased catalyst dosage. Since C-FTB has a lower specific surface area (27.43 m²/g) compared to FTB, the availability of active sites is the limiting factor at low dosages. Consequently, increasing the catalyst mass compensates for this limitation by increasing the total number of active sites available for the reaction, thereby significantly enhancing its degradation efficiency. Meanwhile, C-TB consistently lags behind both composites across all tested amounts, likely due to its inferior structural characteristics. These findings highlight the importance of optimizing catalyst dosage based on the specific surface properties of the material.

Fig. 11.

Impact of Different Catalyst Amounts on Ciprofloxacin Degradation Efficiency: (a) 5 mg, (b) 20 mg, (c) 40 mg, with Catalysts C-TB, FTB, and C-FTB.

Effect of different initial pH

The degradation efficiency of ciprofloxacin using C-TB, FTB, and C-FTB catalysts was evaluated across different pH levels to determine the optimal conditions for photocatalytic activity. The natural pH of the ciprofloxacin solution was measured at 5.8, which served as a reference point for neutral conditions. Additionally, experiments were conducted at pH levels of 3 (acidic) and 10 (basic) to observe the catalysts’ performance in varying acidic and alkaline environments.

As shown in Fig. 12a, the highest degradation efficiencies were observed at pH 3, where the acidic conditions enhanced the activity of all three catalysts, with efficiencies approaching 95%. This elevated efficiency can be attributed to the protonation of active sites in the acidic medium, which facilitates greater interaction between the ciprofloxacin molecules and the catalyst surfaces⁵². The increased availability of protons likely improves the interaction of ciprofloxacin with catalyst surfaces, thereby boosting the degradation rate under acidic conditions⁵³. This suggests that the catalytic reactions at the active sites of C-TB, FTB, and C-FTB are more favorable in an acidic environment, leading to a significant boost in degradation efficiency.

Fig. 12.

Impact of Different Initial pH Levels on Ciprofloxacin Degradation Efficiency: (a) pH 3, (b) pH Neutral (5.8), (c) pH 10, with Catalysts C-TB, FTB, and C-FTB.

In contrast, at neutral and basic pH levels, especially at pH 10, the degradation efficiency of all catalysts decreased notably (see Fig. 12b and c). Under these conditions, fewer protons are available, which might reduce the interaction between ciprofloxacin molecules and the catalyst surfaces, consequently lowering the overall degradation rate⁵⁴. This trend indicates that C-TB, FTB, and C-FTB are more effective under acidic conditions, where the reaction kinetics are enhanced, resulting in more efficient degradation of ciprofloxacin. These findings highlight the importance of pH as a critical parameter in optimizing photocatalytic degradation processes, with acidic conditions offering a distinct advantage for maximum catalyst performance⁵⁵. Overall, contour plots for pairwise combinations of experimental parameters were generated and presented in Fig. 13 in order to better visualize the independent variables and dependent variable relationship.

Fig. 13.

Contour plots for pairwise combinations of experimental parameters.

Isotherm studies

The graphical demonstration of isotherm models was given in Fig. 14 with linear and nonlinear fitting plots for CIP photodegradation. Based on the obtained plots, correlation coefficients and model parameters of Langmuir, Freundlich and Temkin isotherms were evaluated and summarized in Table 2. According to evaluated data, C-TB composite better fitted to Freundlich while FTB and C-FTB composites better fitted to Langmuir isotherm models with maximum R² values of 0.9782, 0.8899 and 0.9878. The q_m values from Langmuir model were obtained as 29.45, 88.7 and 64.59 mg/g for C-TB, FTB and C-FTB respectively, indicating FTB and C-FTB distinguished as the composites with superior photodegradation capacity. In addition, n values from Langmuir model of all composites were obtained between 1 and 10, confirming the photodegradation of CIP was favorable. Isotherm model parameters derived from linear fitting and linear fitting plots were also presented in Table S3 and Fig. S1.

Fig. 14.

Nonlinear fitting plots of Langmuir, Freundlich and Temkin isotherm models (pH: 5.8, photocatalyst dosage: 20 mg, contact time: 180 min).

Table 2.

Isotherm model parameters for the photocatalytic degradation of CIP according to non-linear fitting plots.

Model	Coefficient	C-TB	FTB	C-FTB
Langmuir	KL (L/mg)	1.59	0.4987	1.0427
	R2	0.9460	0.8899	0.9878
	qm (mg/g)	29.45	88.7	64.59
	RL	0.0155	0.0477	0.0234
Freundlich	KF (mg/g)	16.05	29.81	29.72
	R2	0.9782	0.8505	0.9199
	n	4.7438	2.5126	3.643
Temkin	KT (L/g)	46.23	10.21	21.95
	R2	0.9703	0.8143	0.9668
	b (kJ/mol)	0.5670	0.1651	0.2340

Kinetic studies

The kinetic studies of each of the C-TB, FTB and C-FTB composites were carried out separately and the results are given in Fig. 15. All composites were examined according to the L-H zero order (Fig. 15a, c and e) and pseudo first order (Fig. 15b, d and f) models. While the zero order model represents the cases where the reaction rate is independent of the concentration, the pseudo first order model is more compatible with the cases where the reaction depends on the initial conditions. The experiments were carried out by keeping all other conditions constant except the composite type for all three composites. When the graphs given in Fig. 15 are examined for all three composites, the higher R₂ value was reached in the pseudo first order case. In this case, it can be interpreted that the rate changes depending on the decrease in concentration over time for C-TB, FTB and C-FTB catalysts. While R₂ = 0.9548 for the pseudo first order model of the C-TB catalyst, R₂ = 0.9277 for the zero order. These correlations indicate that C-TB complies with pseudo first order kinetics, but it can also move according to the zero order model and exhibit a constant rate at a certain rate. The difference can be observed more clearly in the FTB catalyst. FTB provided a much stronger fit to the pseudo first order model (R₂ = 0.9595) compared to the zero order model (R₂ = 0.8514). This showed that, unlike the other catalysts, the reaction rate of FTB is more dependent on the concentration. A similar situation to that seen in the C-TB catalyst is also encountered for C-FTB. The C-FTB catalyst also showed a higher fit to the pseudo first order model (R₂: 0.9793) compared to the zero order model (R₂ = 0.9437). Using the slope in the pseudo first order plots, k_app values ​​for C-TB, FTB and C-FTB composites were calculated as 0.0025 min-1, 0.0086 min^− 1 and 0.0057 min^− 1, respectively.

Fig. 15.

Kinetic studies of C-TB, FTB, and C-FTB composites based on L-H zero order and pseudo first order models.

In addition to the results obtained from all these linearized equations, the nonlinear form of the L–H kinetic model was also tested using the initial rates (r, dC/dt) obtained for different initial concentrations. In this context, the r values ​​calculated in the range of C₀ = 5–40 mg L⁻¹ were applied to the nonlinear equation of the L–H model (Eq. 14). Model parameters were determined for all three catalysts, and the fit coefficients (R²) were obtained. The parameters and R2 values ​​obtained from the nonlinear model are given in Table S6 in the SI file. The graph for the experimental and nonlinear L–H model is also shown in Fig. S3 in the SI file. These results showed that the reaction rate was controlled by surface adsorption and that the L–H mechanism was valid for the system.

Scavengers effect

To elucidate the photocatalytic mechanism and identify the primary reactive species responsible for Ciprofloxacin (CIP) degradation, radical trapping experiments were conducted using specific scavengers. Benzoquinone (BQ), isopropyl alcohol (IPA), and ethylenediaminetetraacetic acid (EDTA) were introduced into the reaction system to scavenge superoxide radicals (), hydroxyl radicals (·OH), and photogenerated holes (h⁺), respectively.

The results of the scavenger experiments for FTB, C-TB, and C-FTB composites are presented in Fig. 16. It is clearly observed that the degradation efficiency of CIP was inhibited in the presence of all scavengers compared to the control group, confirming the participation of multiple active species. However, the degree of inhibition varied significantly. The addition of BQ resulted in the most substantial decrease in degradation efficiency for the FTB catalyst (dropping to ~ 68%), indicating that superoxide radicals are the dominant active species driving the photodegradation process⁵².

Fig. 16.

Effect of different scavengers (BQ, IPA, EDTA, and H₂O₂) on the photocatalytic degradation efficiency of CIP over FTB, C-TB, and C-FTB composites.

The inhibition observed with IPA and EDTA suggests that hydroxyl radicals (·OH) and photogenerated holes (h⁺) also play a contributory role in the degradation of CIP, although their impact is secondary compared to superoxide radicals. This finding aligns with the band structure of BaTiO₃, where photo-excited electrons in the conduction band reduce dissolved oxygen to form , which subsequently attacks the antibiotic molecules⁴⁸. The slight increase in degradation in the presence of H₂O₂ can be attributed to its role as an additional source of hydroxyl radicals or an electron acceptor, further promoting the oxidation process.

Proposed photocatalytic mechanism for CIP degradation

The photocatalytic degradation mechanism of ciprofloxacin (CIP) by synthesized tannin-modified BaTiO₃-based catalysts, the overall mechanism via light-induced charge generation, reactive oxygen species (ROS) formation and subsequent CIP oxidation pathways, is schematically shown in Scheme 1.

Scheme 1.

Schematic illustration of the proposed photocatalytic mechanism for CIP degradation over FTB, C-TB, and C-FTB composites under visible light irradiation, showing the charge separation and ROS generation pathways.

Photocatalytic degradation of ciprofloxacin (CIP) by tannin-modified BaTiO₃ composites proceeds primarily via visible-light-activated electron-hole separation and subsequent formation of reactive oxygen species. Upon visible light irradiation, electrons in the valence band (VB) of BaTiO₃ are elevated to the conduction band (CB) (e_CB⁻), leaving behind positively charged holes (h_VB⁺). This can be expressed as follows (Eq. 22):

22

These e_CB⁻ and h_VB⁺ active sites contribute to the formation of reactive radicals (hydroxyl radicals (·OH) and superoxide radicals (·O₂⁻)) (Eqs. 23–26):

23

24

25

26

These active species participate in the photodegradation of CIP (Eqs. 27–29):

27

28

29

CIP undergoes a series of oxidation reactions, including oxidation and opening of the piperazine ring, defluorination, decarboxylation, and C–N/C–C cleavage. These steps ultimately lead to the formation of small organic fragments and subsequent mineralization.

Tannin can inhibit the recombination of electron-hole pairs during photodegradation, act as an electron trap, and act as a carrier. This, in turn, improves the photocatalytic efficiency of the semiconductor. Additionally, tannin modification narrows the band gap (2.7 eV for C-TB and C-FTB; 3.1 eV for FTB), increases visible light absorption, and facilitates charge transfer due to its phenolic/aromatic functional groups^14,56,57.

Reusability studies

The stability and reusability of a photocatalyst are critical parameters for its practical application and economic feasibility. To evaluate the long-term performance of the synthesized composites, recycling experiments were conducted for the degradation of CIP over five consecutive cycles. After each cycle, the catalyst was recovered by centrifugation, washed thoroughly with deionized water and ethanol to remove accumulated organic residues, and dried before the next run.

The reusability results for FTB, C-TB, and C-FTB are presented in Fig. 17. As observed, all composites exhibited a gradual decrease in photocatalytic efficiency over the five cycles. For the FTB catalyst, the degradation efficiency decreased from ~ 94% to ~ 65% in the fifth cycle. Similarly, C-FTB showed a decline from ~ 95% to ~ 68%, while C-TB dropped from ~ 78% to ~ 55%. This reduction in activity can be attributed to several factors: (i) the inevitable loss of catalyst mass during the recovery (centrifugation/washing) steps, (ii) the accumulation of recalcitrant degradation by-products on the catalyst surface, which blocks active sites and shields visible light absorption⁵⁸, and (iii) potential minor photocorrosion or aggregation of.

Fig. 17.

Reusability cycles of FTB, C-TB, and C-FTB composites for the photocatalytic degradation of CIP over 5 consecutive runs.

particles upon repeated use. Despite the observed decline, the catalysts retained more than 50% of their activity after five cycles, suggesting they possess reasonable stability for potential wastewater treatment applications.

ANN model evaluation

Table 3 presents the calculated MSE and R² values for various number of neurons at hidden layer. The optimal topology of the obtained ANN was identified as 5-6-1 (Fig. 18) based on the highest R² values of 0.99078 and 0.99722 for training and testing respectively, together with acceptable MSE values. Regression plots of ANN model outputs from training, testing and validation data revealed a good relationship with the experimental data sample with R² values close to 1 (Fig. 19a). ANN studies on CIP removal by photodegradation have been reported previously with photocatalysts including TiO₂ nanoparticles²⁹ and TiO₂ nanoparticles on montmorillonite, R² value of 0.9757 with 4-5-1 configuration and 0.9864 with 4-14-1 configuration were obtained respectively. As observed from the Fig. 19a, overall data with a R² value of 0.99118 indicating a well performed ANN configuration. Although normalized experimental data vs. normalized predicted data (Fig. 19b) and MSE values (Fig. 19c) demonstrated a slight scattering of data, overall performance of the developed ANN was found to be satisfying for the removal of CIP by photocatalytic degradation. The mean absolute error (MAE = 0.0465) demonstrates that, on average, the model’s predictions deviate from the true values by less than 0.05 on the normalized scale (Table 4). Final architecture of the designed ANN was presented in Fig. 20.

Table 3.

MSE and R² values for various number of neurons at hidden layer.

Neurons	Trian MSE	Train R2	Test MSE	Test R2
1	0.0828	0.83227	0.11	0.77957
2	0.0481	0.90482	0.089	0.81027
3	0.0385	0.91765	0.14	0.81066
4	0.00551	0.9868	0.024	0.97396
5	0.00505	0.98601	0.008	0.98303
6	0.000802	0.99078	0.18	0.99722
7	0.000821	0.98346	0.031	0.94385
8	0.000386	0.96101	0.2	0.76733
9	0.000104	0.9958	0.15	0.76289
10	0.000194	0.99575	0.2	0.90069
11	0.000258	0.99193	0.031	0.95235
12	0.000258	0.98024	0.59	0.88521
13	0.00000052	0.99986	0.12	0.81681
14	0.00000068	0.96466	0.21	0.89629
15	0.000194	0.99953	0.81	0.98271
16	0.000257	0.99832	0.86	0.85605
17	0.000194	0.99948	0.14	0.86404
18	0.000259	0.9350	0.15	0.96311
19	0.00000051	0.94359	0.032	0.95745
20	0.0000011	0.99999	0.044	0.86339

Fig. 18.

Schematic demonstration of optimum ANN configuration (5-6-1).

Fig. 19.

(a) Regression plots of ANN model outputs, (b) Normalized experimental data vs. normalized predicted data, (c) Output error values of samples, (d) Residual vs. normalized predicted data, (e) Q-Q Plot of normalized experimental data vs. normalized predicted data quantiles, (f) Cross-Validation error plot (Fig. S2).

Table 4.

Performance metrics for ANN.

Parameter	Value
R2	0.991
R2adj	0.975
MSE	0.0046
MAE	0.0465
SEP (%)	15.16
MRPD (%)	4.50

Fig. 20.

Architecture of artificial neural network designed for the prediction of ciprofloxacin removal.

Table S4 shows a list of input weights connected the input layer to hidden layer and hidden layer to output layer as represented in Fig. 20, together with biases connected to hidden and output layer. Weights and biases were further used for the sensitivity calculations by Garson algorithm. The values derived in each step of the Garson algorithm were presented in Table S5 and Table 5. Considering the relative importance of all parameters (Fig. 21), the significant effect of photocatalyst type was confirmed with the highest percentage of 31.4% followed by initial concentration of CIP. Displaying importance values in the range of 15–20%, pH and photocatalyst dosage parameters were also determined to have a notable effect on CIP removal. However, contact time was found to be the least effective parameter with an importance value of 4.3%.

Table 5.

Sensitivity analysis results.

Sum R	Parameters	Relative % importance
6.25838	Photocatalyst type	31.4
0.855866	Contact Time (min)	4.3
5.58739	Initial concentration of CIP	28.0
3.321934	pH	16.7
3.911488	Photocatalyst dosage (mg)	19.6

Fig. 21.

Relative % importance of input parameters.

Conclusion

This study successfully demonstrated that modifying barium titanate with tannin, a natural biopolymer, is a highly effective strategy for developing sustainable visible-light photocatalysts. Among the synthesized composites, the formaldehyde-tannin modified BaTiO₃ (FTB) exhibited superior photocatalytic performance for ciprofloxacin degradation, significantly outperforming its calcined counterparts. The enhanced activity of FTB is primarily attributed to its unique synthesis route; avoiding calcination preserved the polymeric tannin network, resulting in an amorphous, spongy morphology with an exceptionally high specific surface area. This structural advantage, combined with a reduced band gap, facilitated efficient charge separation and provided abundant active sites for the adsorption and degradation of antibiotic molecules.

Furthermore, the experimental findings were corroborated by Artificial Neural Network (ANN) modeling, which demonstrated high predictive accuracy (R² > 0.99) and identified the catalyst type as the most influential parameter affecting the process. The consistency between the kinetic studies, which followed pseudo-first-order kinetics, and the ANN predictions highlights the robustness of the proposed system.

From an environmental perspective, utilizing a low-cost, biomass-derived modifier like tannin offers a promising, eco-friendly alternative to expensive doping agents for wastewater treatment. However, to transition this technology from laboratory scale to industrial application, further research is essential. Future studies should prioritize evaluating the catalyst’s long-term stability and regeneration capacity, testing its performance in real wastewater matrices containing complex organic and inorganic interferences, and designing continuous-flow reactor systems for large-scale operations.

Supplementary Information

Below is the link to the electronic supplementary material.

Author contributions

**S.S., T.H.S. and E.K.:** conceptualization, methodology, investigation, writing- original draft, writing – review & editing. **N.E.A.:** conceptualization, methodology, writing – review & editing. **G.N.S.:** conceptualization, methodology, writing – review & editing, supervision.

Funding

This support covered material purchases and analysis expenses. No additional financial support was received.

Data availability

The datasets generated and/or analyzed during the current study are available from the corresponding author upon reasonable request.

Declarations

Competing interests

The authors declare no competing interests.