Application of Advanced Ceratophyllum demersum Activated Carbon in Toxic Dye Adsorption: Optimization via Box–Behnken Design and Artificial Neural Network

ABSTRACT

Water contamination by cationic dyes such as crystal violet (CV) is a significant environmental concern due to their toxicity and persistence. This study investigated and optimized CV adsorption onto Ceratophyllum demersum activated carbon (CDAC) using the Box–Behnken design (BBD) and artificial neural network (ANN) models. CDAC exhibited a specific surface area of 190.56 m² g⁻¹ and a mesoporous structure containing oxygen‐ and phosphorus‐bearing functional groups. Batch adsorption experiments showed adsorption capacities ranging from 95 to 290 mg g⁻¹, with adsorption following pseudo‐second‐order kinetics and exhibiting favorable isotherm behavior. Although the ANN model produced lower prediction errors (≤ 8%) than the BBD model (≥ 20%), the BBD model was considered more reliable given the limited experimental dataset. Optimization predicted a maximum adsorption capacity of 258.76 mg g⁻¹ and a removal efficiency of 86% at an initial CV concentration of 150 mg L⁻¹, a CDAC dose of 0.5 g L⁻¹, and a contact time of 10 min. Germination assays using Lens culinaris seeds indicated reduced phytotoxicity after treatment. These results demonstrate that CDAC is an effective and sustainable adsorbent for CV removal from aqueous solutions.

Summary

• CV adsorption on CDAC was modeled and optimized using ANN and BBD chemometric tools.

• CDAC achieved 258.76 mg g⁻¹ of CV adsorption capacity under desirability function optimal [CV] = 150 mg L⁻¹ and CDAC dose = 0.5 g L⁻¹.

• CDAC maintained up to 180.72 and 182.26 mg g⁻¹ of CV adsorption capacity after five thermal and distilled water regenerations, respectively.

Ceratophyllum demersum activated carbon was prepared and characterized (SEM–EDX, BET, FTIR) and demonstrated efficient crystal violet removal via endothermic physisorption following pseudo‐second‐order kinetics and the Freundlich isotherm. ANN modeling and Box–Behnken optimization confirmed its performance, along with stability over more than five regeneration cycles and no observed ecotoxicity.

1. Introduction

The permanence, toxicity, and mutagenic potential of industrial dyes pose a serious environmental problem due to water contamination (Manyani et al. 2026). Among these pollutants, crystal violet (CV), a triphenylmethane cationic dye, is particularly difficult to remove from wastewater due to its high solubility and chemical stability. CV in water destroys fauna and flora and requires the development of processes and materials for its removal. Adsorption on activated biochar is an easy‐to‐use process for decentralized applications or areas with limited technical capacity (Oghoghorie et al. 2025). Activated biochar exhibits a high capacity for removing CV dye, even at low concentrations, owing to its large specific surface area, diverse surface functional groups, and micro‐ to medium‐porous structure. L. Tang et al. (2025) demonstrated that raising the thermal decomposition temperature to 800°C resulted in biochar with a high specific surface area and a microporous structure, which directly impacted its CV adsorption efficiency. Micro‐ and mesoporosity contributed to an increased number of available active sites, whereas the surface functional groups, in conjunction with the sodium alginate matrix, enhanced the strong interaction with the dye molecules, even at low dye concentrations (Ighnih et al. 2026; Ziane et al. 2026). The adsorption on activated biochar is spontaneous, stable, and efficient (L. Tang et al. 2025). G. Tang et al. (2024) assumed that alkaline modification of pomelo peel biochar improved the pore structure and increased the diversity of surface functional groups and enabled multiple interactions with the CV dye, including pore filling, hydrogen bonding, π–π interactions, and electrostatic attraction, resulting in high adsorption capacities. Raji et al. (2023) demonstrated that Moringa oleifera residue‐activated carbon exhibited a superior specific surface area and rich porous structure, reflected in high CV adsorption capacity.

Economically, activated biochar is a relatively inexpensive option, especially when produced from local biomass in accordance with circular economy principles (Subudhi et al. 2025). Adsorption does not produce intermediate products that are potentially more toxic than the original pigment, which is a known problem with some advanced oxidation processes (Hazarika et al. 2025). Environmentally, activated biochar is a relatively eco‐friendly, low‐toxicity, and regenerable material by simple washing or heat treatment without any potential risk (Zafeer et al. 2025). Also, it is characterized by high porosity, large specific surface area, and diverse functional groups that promote cationic molecule adsorption (Khandaker et al. 2025).

Ceratophyllum demersum (CD) is a macrophyte that constitutes a biomass that is easy to harvest because it is very common in freshwater environments, forming dense stands that are often considered plant waste but could survive and remove organic pollutants from aqueous media (Ali, Ali, et al. 2023, 2026; Kumar et al. 2025; Wang et al. 2025).

This study prepares activated carbon derived from CD for the adsorption of CV following characterization by FTIR, SEM–EDX, Brunauer–Emmett–Teller (BET), and pH_pzc. Adsorption is studied as a function of pH, adsorbent dose, dye concentration, contact time, temperature, and ionic strength. This process is then optimized by combining BBD with ANN. The originality and novelty of this study lie in the combination of experimental optimization using BBD with ANN modeling to analyze and compare predictive capabilities regarding the CV dye adsorption onto activated carbon derived from CD, an aquatic plant and biomass that is abundant and easy to harvest from aquatic media. In addition, it explores the relationship between adsorbent structural properties and the adsorption behavior of CV dye molecules. This paper aims to identify complex interactions between experimental variables and determine optimal adsorption capacity. Thus, this study proposes an ecological and efficient solution for the depollution of water loaded with dyes while valorizing local and renewable aquatic biomass.

2. Materials and Methods

2.1. Preparation and Characterization of CD Activated Carbon

The aquatic plant CD was collected from a mountain lake in the High Atlas Mountains of Morocco in the Ifrane region. The CD was initially air‐dried, ground, and sieved to obtain a fine powder of < 80 μm. A 20 g sample of CD was impregnated with 40 mL of H₃PO₄ (85%) for 20 min. The impregnated CD was washed several times with distilled water to remove residual impurities until a neutral pH was achieved. After filtration and drying at 120°C in an oven for 24 h, the material was calcined in an IBX LBX MF12 muffle furnace at 300°C for 2 h under anoxic conditions, in the absence of oxygen, to obtain partially carbonized biochar, then cooled and ground in a porcelain mortar into smaller particles. Figure 1 summarizes the steps involved in CDAC preparation.

FIGURE 1.

CDAC preparation experimental protocol.

The CDAC adsorbent was characterized using a set of complementary analytical techniques to determine its functional groups, morphology, porosity, and surface charge, which are essential for understanding its interaction with CV. FTIR spectra were recorded using an FTIR‐VERTEX 70 spectrometer (Bruker Optics Inc., Billerica, USA) in the 4000–400 cm⁻¹ range to identify the main functional groups likely to participate in chemical interactions with the dye. This technology is based on selective infrared radiation absorption by chemical bonds that induces vibrational transitions (Al‐Amin et al. 2025). CDAC morphology was analyzed using the FEI Quanta 200 SEM–EDX manufactured by FEI in Hillsboro, Oregon, United States. This method relies on the interaction of a focused electron beam with the sample surface, producing various electronic signals that enable a 3D morphology (Ali, Zhang, and Santos 2023). The specific surface area, pore volume, and pore size were determined using the BET method through the JWGB 200 analyzer. Furthermore, the surface charge of the CDAC was evaluated by determining the point of zero charge (pH_pzc) to predict the electrostatic behavior of the material as a function of the pH by introducing 0.1 g of CDAC into 20 mL of 0.05 M NaCl over a pH range of 2–12, adjusted with 0.1 M HCl/NaOH. Final pH measurements were measured after filtration (0.45 μm), and the pH_pzc was determined as the initial pH at which ΔpH (final pH − initial pH) = 0.

2.2. Targeted Adsorbate

CV is a basic cationic dye listed under the generic name Basic Violet 3. Its molecular formula is C₂₅H₃₀N₃Cl, with a molar mass of 407.98 g mol⁻¹. Its aqueous solution has a pH of approximately 5.9 and exhibits maximum absorption in the UV–visible range at 586 nm. The products used had an analytical purity of ≥ 90% and were provided by Sigma‐Aldrich (Germany). Structurally, CV is a triamine aromatic molecule composed of a triphenylmethane ring substituted with amine groups, which confer the dye its cationic charge and high affinity for negatively charged surfaces. Although widely used in industrial applications, it is toxic at high doses and potentially mutagenic; hence, the importance of its effective removal from aqueous media is crucial. The stock solution prepared for the experimental tests was obtained by dissolving and magnetically stirring 1 g of CV powder in distilled water (1 L).

2.3. Adsorption Experimental Protocol

Adsorption tests were carried out in batch mode using CV solutions (0.2 L) prepared with initial concentrations between 50 and 150 mg L⁻¹, and then brought into contact with doses of CDAC adsorbent ranging from 0.5 to 2 g L⁻¹. The mixtures were stirred for a specific time to facilitate adsorption and then centrifuged at 6000 rpm for 3 min to separate the adsorbent from the liquid medium. Adsorption kinetics were studied over contact times ranging from 5 to 180 min, pH values between 2 and 12, and temperatures between 20°C and 60°C, considering the presence of salts such as NaCl, KCl, CaCl₂, and MgCl₂. The absorbance relative to the residual CV concentration was measured by ultraviolet–visible spectrophotometry at 586 nm and is used to calculate the adsorption capacities at time t.

2.4. Chemometric Methodology for Modeling and Optimizing CV Adsorption

The chemometric approach combines a Box–Behnken design (BBD) with a polynomial model and a single‐hidden‐layer artificial neural network (ANN) to analyze the adsorption of CV on CDAC, considering contact time, CV concentration, and CDAC dose. Performance is verified using R ², RMSE, MAE, residual analysis, and Q–Q plots. Response surfaces and network profiles describe the interactions between variables and confirm the observed trends. Finally, optimization was performed using the desirability function applied to the BBD model to identify the operating conditions that led to maximum adsorption capacity.

2.5. Lens culinaris Seeds Germination Experimental Protocol

The phytotoxicity of CV was assessed before and after adsorption using uniformly sized Lens culinaris seeds incubated for 8 days in Petri dishes containing either an untreated CV solution at 150 mg L⁻¹ or the same solution after adsorption treatment. The incubation conditions were strictly controlled, and three biological parameters were monitored: seed germination rate, root length, and stem length. Comparing these indicators between the treated and untreated solutions allowed for the assessment of residual CV toxicity after removal via adsorption.

3. Results and Discussion

3.1. Characterization of CDAC

Figure 2 shows the results of the characterization of the CDAC by FTIR, SEM–EDX, and BET. The FTIR spectrum revealed various chemical functional groups on the CDAC surface, indicating its high surface activity. The broad band extending between 3600 and 3200 cm⁻¹ is attributed to the stretching vibrations of hydroxyl groups, whether from absorbed moisture or phenolic groups, which play a key role in the formation of hydrogen bonds with contaminants (Pasieczna‐Patkowska et al. 2025). The signal at approximately 2754 cm⁻¹ was due to aliphatic C–H vibrations, indicating the presence of organic compounds of plant origin (Tsaousis et al. 2025). The bands at 2302 and 2121 cm⁻¹ are often attributed to CO₂ absorbed into the pores or weak triple‐bond vibrations, indicating an open‐pore structure. The significant peak at 1603 cm⁻¹ reflects the C=C vibrations of the aromatic rings or the C=O vibrations of the kinin structures, confirming the progression of carbonization and the formation of the aromatic skeleton of activated carbon (Aravind and Kamaraj 2024). The band at 957 cm⁻¹ is associated with the C–O vibrations of alcohols or ethers, which are active sites for surface interactions, whereas the signal at 462 cm⁻¹ indicates Si–O vibrations or metal–oxygen bonds from the natural mineral ash of the plant (Azeez and Shenbagaraman 2025).

FIGURE 2.

Characterization of CDAC by FTIR, SEM–EDX, and BET.

The SEM images show a highly porous morphology consisting of irregular sheets and well‐developed cavities, indicating efficient opening of the plant matrix and a significant increase in the accessible surface area of the adsorbent. The EDX spectrum reveals the elemental composition of the activated carbon. An intense peak at approximately 0.28 keV, attributed to carbon, confirmed that carbon was the main component of the material (Shittu and Pwavodi 2025). A peak at approximately 0.53 keV was attributed to oxygen, which is consistent with the FTIR results and confirms the presence of oxygenated surface groups, such as ‐OH, C=O, and C–O (Aghabekyan et al. 2025). The peaks at 1.75 and 2.02 keV are generally attributed to phosphorus from phosphoric acid (Ahmad and Zubair Ansari 2025).

BET analysis revealed a specific surface area of 190.56 m² g⁻¹, total pore volume of 0.266 cm³ g⁻¹, and mean pore diameter of 5.58 nm, indicating the predominance of mesopores formed during chemical activation.

3.2. One Parameter at a Time Study

3.2.1. Effect of Contact Time

Figure 3a,b shows that adsorption increases rapidly during the first few minutes and then stabilizes, indicating a two‐phase kinetic: rapid surface adsorption followed by slower intraparticle diffusion. Abo‐Mousa et al. (2025) highlighted the crucial role of contact time in controlling the adsorption kinetics of alizarin Red S dye on unmodified and aluminum oxide‐modified white bean straw. Their results show rapid adsorption during the first 15 min due to the high availability of active sites. The rate then decreases as the surface becomes saturated, until equilibrium is reached (Abo‐Mousa et al. 2025). Chutani et al. (2026) similarly observe that the adsorption efficiency of methylene blue on the surface of tamarind shells increases rapidly at first, then slows down after the 4th hour, with equilibrium reached around the 8th hour. These behaviors indicate that adsorption is initially governed by surface mass transfer mechanisms; however, diffusion within the pores becomes the limiting factor over time (Oyehan et al. 2026).

FIGURE 3.

Kinetic evaluation of adsorption parameters.

3.2.2. Effect of CV Concentration

Figure 3a shows an increase in adsorption capacity from 95.45 to 290.23 mg/g as the initial dye concentration increases from 50 to 150 mg/L. This increase results from a higher flux of molecules toward the active sites, despite increased competition upon saturation. Muteeb et al. (2025) also demonstrate the significant influence of the initial concentration of methylene blue on adsorption by activated chicken bones. In the range of 10–50 mg/L, removal remains high due to a large number of available sites and rapid transfer of molecules to the adsorbent surface (Muteeb et al. 2025). Admawi (2025) observes a similar trend for quinoline yellow adsorbed by bitter orange residues. Adsorption initially increases with concentration, then plateaus when the active sites become saturated, and competition between molecules limits further binding (Admawi 2025). El Bourachdi et al. (2026) analyzed the effect of the initial dye concentration on the adsorption of Blue 25 by cellulose extracted from Aloe vera . Increasing the initial concentration gradually decreased the removal efficiencies for both dyes. This behavior is explained by the saturation of active adsorption sites on the cellulose surface as the number of dye molecules in solution increases, which promotes saturation and competition for the binding sites. The removal efficiency showed a sharp decline due to the weak interactions between the dye and the cellulose surface, which are dominated by limited physical forces (El Bourachdi et al. 2026).

3.2.3. Effect of CDAC Dose

Figure 3b shows a decrease in adsorption capacity from 287.82 to 73 mg g⁻¹ as the adsorbent dose increases from 0.5 to 2 g L⁻¹, and this could be attributed to particle agglomeration and active sites hiding. Peighambardoust et al. (2025) also observe that a higher dose improves the overall efficiency of CV removal, reaching 95%, due to the higher number of available active sites. They showed that exceeding certain dosages did not lead to any further significant improvement, as the adsorption efficiency became almost constant, indicating that the system reached a state of relative saturation of the active sites (Peighambardoust et al. 2025).

3.2.4. Effect of Solution Temperature

Adsorption increased between 20°C and 40°C from 252.27 to 287.74 mg g⁻¹, suggesting an endothermic process: The thermal energy input promotes the adsorbent–adsorbate interaction, either by increasing molecular agitation or by activating new sites of interaction (Figure 3c). Above 40°C, the variation becomes small, indicating that the system reaches a quasi‐plateau zone where the adsorption mechanisms are balanced by slight thermal desorption. This indicates that the adsorption energy is neither too low (otherwise, heat would inhibit adsorption) nor too high (otherwise, the temperature would have little effect on the adsorption). Yildiz et al. (2026) describe the exothermic adsorption of the CV dye on Trichoderma reesei. An increase in temperature reduces removal efficiency, and low enthalpy values indicate physical adsorption based on noncovalent interactions (Yildiz et al. 2026). In contrast, Sun et al. (2026) observe a higher CV adsorption rate on a lignosulfonate‐modified magnesium–aluminum layered double hydroxide composite as the temperature rises because higher temperatures facilitate access to the active sites of the porous structure.

3.2.5. Effect of Ionic Strength

Figure 3d shows that the presence of electrolytes changed the adsorption capacity, showing the importance of solution ionic strength. Monovalent sodium and potassium cations and divalent calcium and magnesium cations modulate adsorption based on their concentration and charge. Divalent ions have a stronger effect, likely because they compress the electrical double layer around the adsorbent or form a complex with the dye, which affects CDAC–CV interaction. At 0.3 M, there was a big drop overall, which could be due to ionic competition for active sites or an electrostatic shield that made the charged dye less attracted to the adsorbent surface. Cetina et al. (2025) observed higher adsorption of Reactive Black 5 on activated carbon in a saline ionic environment. Sodium chloride or sodium sulfate up to 1 M enhances electrostatic interactions, hydrogen bonding, and π–π interactions (Cetina et al. 2025). Chadha and Berry (2025) obtained the opposite result for the CV dye using papaya peels. Increasing the sodium chloride concentration from 0.01 to 0.5 M caused the removal efficiency to drop from 67.53% to 43.94% because sodium ions compete with CV cations for adsorption sites (Chadha and Berry 2025). Nouioua et al. (2025) report a minor impact of horn biochar and show that increasing the sodium chloride concentration from 0.05 to 1 M reduces the adsorption capacity by only about 2.3%.

3.2.6. Effect of Solution pH

Figure 3e shows a clear effect of pH on adsorption. In an acidic environment, the protonated surface limits the binding of the cationic dye (Equation 1). When the pH rises to 10, the surface becomes negative and adsorption increases (Equations 2 and 3). Beyond that, the equilibrium remains stable because the active sites reach saturation and hydroxyl anions compete for binding sites (Equation 4). Figure 3f indicates a pH_pzc of 6.38. Below this value, the surface carries a positive charge. Above it, the surface becomes negative, which enhances the attraction of cationic dyes (Chadha and Berry 2025; Dalal 2025; Edokpayi 2025; Kouda et al. 2025; Usman et al. 2025).

$$\mathrm{S}–\mathrm{OH}+{\mathrm{H}}^{+}\to \mathrm{S}–{\mathrm{OH}}_2^{+}$$ (1)

$$\mathrm{S}–\mathrm{OH}\to \mathrm{S}–{\mathrm{O}}^{-}+{\mathrm{H}}^{+}$$ (2)

$$\mathrm{S}–{\mathrm{O}}^{-}+{\mathrm{CV}}^{+}\to \mathrm{S}–{\mathrm{O}}^{-}\cdots {\mathrm{CV}}^{+}$$ (3)

$$\mathrm{S}–\mathrm{OH}+{\mathrm{OH}}^{-}\to \mathrm{S}–{\mathrm{O}}^{-}+{\mathrm{H}}_2\mathrm{O}$$ (4)

3.3. Isotherms, Kinetic Models, and Thermodynamics

Table 1 shows that adsorption follows pseudo‐second‐order kinetics with R ² ₂ ∈ [0.997, 0.999]. The intraparticle diffusion constants (K _IPD) also increased from 6.4 to 22.9, indicating that internal diffusion played a role in the overall process. However, R ² _IPD ∈ [0.56, 0.89] suggests that this mechanism is not predominant, confirming that adsorption is primarily governed by surface processes, likely involving specific interactions.

TABLE 1.

Kinetic model parameters.

[CV] (mg L−1)	Q e (mg g−1)	Pseudo‐first‐order		Pseudo‐second‐order		Intraparticle diffusion
		K 1	R 2 1	K 2	R 2 2	K IPD	C IPD	R 2 IPD
50	95.0359	0.04379451	0.8517072	0.0031513027	0.9997026	6.396655	38.28583	0.6299095
75	145.3726	0.05401578	0.9409247	0.0013113225	0.9991869	10.991928	47.48861	0.7274060
100	189.8732	0.04759841	0.8412818	0.0034782627	0.9999242	12.210689	83.78835	0.5624709
125	236.3939	0.03181263	0.9857502	0.0002680198	0.9978309	19.161003	48.62587	0.8985795
150	290.0976	0.05047574	0.9919283	0.0003724347	0.9989610	22.920774	79.28727	0.8180529

Isotherm analysis (Table 2) reveals that no model perfectly describes the system, but certain parameters allow us to identify the nature of the adsorption. The Freundlich model (R ² = 0.6712) fit better than the Langmuir model (R ² = 0.2509), suggesting a heterogeneous surface and moderate multilayer adsorption. The parameter n _F = 1.337 > 1 indicates favorable adsorption, consistent with the Langmuir R _L = 0.099, which is also within the favorable adsorption zone. The Dubinin–Radushkevich model provides a low adsorption energy K = 10⁻⁶, consistent with physically activated adsorption, but the theoretical capacity of approximately 281 mg g⁻¹ is in agreement with the experimental Q _e values, confirming a porous surface that offers non‐negligible volumetric adsorption. The Temkin model (R ² = 0.6355) suggests the existence of adsorbent–adsorbate interactions that gradually decrease with coverage, which is typical of energetically varied sites. In Albo Hay Allah and Alshamsi (2023), the activated carbon/zinc oxide nanocomposite derived from Pontederia crassipes acts as an adsorbent for Congo Red and Malachite Green. The adsorption phenomenon is described by the Freundlich model, which describes adsorption on a heterogeneous surface with several types of active sites. Each site has a different adsorption energy. This observation indicates a non‐uniform distribution of pores and surface functions in the AC/ZnO material. The structure of activated carbon provides high microporosity. ZnO nanoparticles add additional active sites linked to hydroxyl groups and surface defects. It is also described by Halsey's isotherm corresponding to multilayer adsorption on a heterogeneous surface. The fit of this model suggests a gradual accumulation of dye molecules in the pores of the material. The adsorption capacity reaches approximately 987.77 mg/g for Congo Red and 74.36 mg/g for Malachite Green. The high capacity for Congo Red can be explained by the electrostatic interactions between the charged surface of the composite and the dye, the π–π interaction between the aromatic rings of the dye and the carbon matrix, and adsorption in the micropores of the activated carbon. Taken together, this indicates a mechanism dominated by multilayer physical adsorption on a heterogeneous surface (Albo Hay Allah and Alshamsi 2023). Salman and Alshamsi (2022) combined activated carbon and sulfonated polystyrene spheres and found that Rhodamine B adsorption is heterogeneous following the Freundlich isotherms and found that sulfonate groups on the surface increase the affinity for cationic species such as Rhodamine B. However, the adsorption capacity differs greatly between the two dyes: 98% removal for Congo Red and 34% for Rhodamine B. This difference is due to several parameters: dye molecular size, molecule charge, and material pore accessibility. Congo Red has a linear structure and fits more easily into the pores. Rhodamine B has a more voluminous structure and diffuses less efficiently into the porous matrix. The Freundlich model, therefore, indicates an energetically heterogeneous surface, multilayer adsorption, and a wide distribution of active sites. The thermodynamic data also show an endothermic and spontaneous process. Increasing the temperature, therefore, improves the diffusion of the dye into the pores (Salman and Alshamsi 2022).

TABLE 2.

Isotherm model parameters.

	Parameter	Value
Langmuir	Q L	666.67
	K L	0.091
	R 2	0.2509
	R L	0.0990099
Freundlich	k F	63.64
	n F	1.337
	R 2	0.6712
Dubinin–Radushkevich	K	0.000001
	Q m	281.519017
	R 2	0.6834
Temkin	B 1	127.2
	K T	0.127
	R 2	0.6355

The thermodynamic parameters (Table 3) corroborate a spontaneous process because ΔG° is negative at all temperatures from −5.7 to −12.1 kJ mol⁻¹, with more favorable adsorption at higher temperatures, indicating an endothermic process consistent with the positive ΔH° = 29.96 kJ mol⁻¹. This value places the adsorption at the interface between enhanced physisorption and slight chemisorption, which corresponds to the kinetic data (Guesmi et al. 2025). The positive entropy ΔS° = 0.129 kJ mol⁻¹ suggests an increase in disorder at the solid–solution interface, probably due to the release of hydrated water molecules and structural reorganization of the adsorbed species (Areti et al. 2025). The overall results indicate favorable endothermic adsorption, governed by surface chemisorption dominated by specific interactions, but assisted by a secondary contribution of intraparticle diffusion and significant heterogeneity of the sites.

TABLE 3.

Thermodynamic study parameters.

Thermodynamic parameters	Temperature (K)	Value
ΔG° (kJ mol−1)	293.15	−5.748
	303.15	−12.102
	313.15	−10.021
	323.15	−11.730
	333.15	−12.093
ΔH° (kJ mol−1)	—	29.961
ΔS° (kJ mol−1)	—	0.129

3.4. Chemometric Modeling and Optimization of CV Adsorption on CDAC

A cross‐analysis of the experimental conditions and adsorption capacities measured in Table 4 highlights the high sensitivity of the CV adsorption system on the surface of CDAC to contact time, initial concentration, and CDAC dose. The high adsorption capacity observed at CV concentrations of 100 and 150 mg/L with a CDAC dose of 0.5 g L⁻¹ indicates that the available specific surface area of the material is rapidly saturated and that the concentration gradient is the primary factor driving diffusion processes. In contrast, the low capacities observed at 2 g L⁻¹ illustrate the effect of agglomeration of the adsorbent particles. The three tests at the center of gravity of the cubic experimental domain were perfectly replicated to confirm the proper control of the experiments and provide a reliable reference basis for BBD models and ANN complementary modeling. Although the ANN model showed lower relative error values (≤ 8%) than the BBD model (≥ 20%) in the dataset studied, it presents a risk of overfitting, as the network was trained on only 15 experimental points. Consequently, ANN is considered here only as a complementary tool for comparison in the experimental design space, whereas the Box–Behnken model remains the most reliable optimization framework for this dataset.

TABLE 4.

Matrix of experiments, responses, and residuals for CV adsorption onto CDAC.

Contact time (min)	CV concentration (mg L−1)	CDAC dose (g L−1)	Measured adsorption capacity (mg g−1)	ANN predicted adsorption capacity (mg g−1)	ANN residuals (mg g−1)	ANN absolute error (mg g−1)	ANN relative error (%)	BBD predicted adsorption capacity (mg g−1)	BBD residuals (mg g−1)	BBD absolute error (mg g−1)	BBD relative error (%)
10	50	1.25	36.186	24.030	12.156	12.156	50.587	28.990	−4.960	4.960	17.109
100	50	1.25	38.130	37.740	0.390	0.390	1.033	28.990	8.750	8.750	30.183
10	150	1.25	106.754	114.070	−7.316	7.316	6.414	122.080	−8.010	8.010	6.561
100	150	1.25	115.290	115.570	−0.280	0.280	0.242	122.080	−6.510	6.510	5.333
10	100	0.50	173.866	189.310	−15.444	15.444	8.158	180.760	8.550	8.550	4.730
100	100	0.50	184.626	187.860	−3.234	3.234	1.721	180.760	7.100	7.100	3.928
10	100	2.00	46.687	42.620	4.067	4.067	9.542	48.070	−5.450	5.450	11.338
100	100	2.00	48.877	59.280	−10.403	10.403	17.549	48.070	11.210	11.210	23.320
55	50	0.50	94.366	101.020	−6.654	6.654	6.587	102.460	−1.440	1.440	1.405
55	150	0.50	270.186	246.940	23.246	23.246	9.414	259.060	−12.120	12.120	4.678
55	50	2.00	23.952	24.500	−0.548	0.548	2.237	33.280	−8.780	8.780	26.382
55	150	2.00	72.767	67.680	5.087	5.087	7.516	62.860	4.820	4.820	7.668
55	100	1.25	77.466	78.110	−0.644	0.644	0.824	75.540	2.570	2.570	3.402
55	100	1.25	77.466	78.110	−0.644	0.644	0.824	75.540	2.570	2.570	3.402
55	100	1.25	77.466	78.110	−0.644	0.644	0.824	75.540	2.570	2.570	3.402

Figure 4 illustrates the ANN connections between the operating parameters and the adsorption capacity of the CV dye by CDAC. The weights associated with the different connections show that the three input variables do not contribute equally; some influences are positive, whereas others are negative, reflecting synergistic or antagonistic effects, depending on the hidden neurons. The dispersion and sign of the weights indicate that some information is strongly compressed by the hidden layer, meaning that the network extracts nonlinear combinations of variables to reproduce adsorption dynamics. R ² ≥ 0.99, MSE ≤ 817.9, RMSE ≤ 23.10, MAE ≥ 19.29, error ≤ 0.01, and iterations = 42 of the ANN model confirmed that its architecture may accurately reflect the combined variations in [CV], contact time, and CDAC dose on the adsorption performance.

FIGURE 4.

ANN architecture for CV Adsorption onto CDAC.

Williams plot (Figure 5) examines the influence of each experimental data point on the ANN model and its residuals distribution. The leverage values remained below the critical threshold, suggesting that no single observation had a disproportionate impact on the ANN model's internal parameters. The absence of systematic anomalies in the forecasts is reflected in the normalized residuals, which are scattered around 0 without any discernible trend or exceeding the approved thresholds. ANN model correctly reproduces central values and the most extreme observations, as evidenced by a few outliers that, nevertheless, remain within the limits. Consequently, the leverage‐residual combination exhibits a stable predictive structure with acceptable chemometric performance and no undesirable outliers (Benjelloun et al. 2026).

FIGURE 5.

ANN William's plot for CV adsorption onto CDAC.

Figure 6 describes the performance of the ANN and BBD models. The ANN residuals remain close to zero, indicating a low and stable error. The BBD model shows more pronounced deviations for certain high adsorption values. The Q–Q plot confirms a more uniform error distribution for the ANN. The comparison of measured versus predicted values yields R ² = 0.9813 for the ANN and 0.9863 for the BBD. The network training stabilizes after approximately 40 iterations. However, the use of 15 trials indicates a risk of overfitting. The Box–Cox transformation improves the homogeneity of variance. Cook's distance analysis indicates a few influential trials associated with high adsorption values without affecting the BBD model.

FIGURE 6.

ANN and BBD diagnostic analysis for CV adsorption onto CDAC.

ANOVA in Table 5 confirms the robustness of the BBD model (F = 46.61, p = 0.0003). The initial dye concentration and the CDAC dose strongly influence adsorption, with F‐values of 116.62 and 236.93. Contact time has little effect within the studied range, indicating rapid equilibrium. Although the AB (p‐value = 0.7977 > 0.05) and AC (p‐value = 0.7395 > 0.05) interactions show negligible contributions, indicating a relative independence between contact time and the other two variables, the bc interaction stands out clearly (p‐value = 0.0017 < 0.05). This synergy between the initial dye concentration and the biosorbent dose indicates that the material's effectiveness depends not only on the amount of active surface area but also on the initial load that this surface must bear (Mahmood et al. 2025). Furthermore, the quadratic terms confirm this dynamic: only the curvature associated with the biosorbent dose (C²) is statistically significant (p‐value = 0.0017 > 0.05), reflecting a nonlinear response as the material dose increases, which is likely linked to the progressive saturation of the active sites. Finally, the absence of experimental error reflected the good reproducibility of the tests, whereas the lack of a significant fit suggested a good fit of the model to the available data. Therefore, the model can be described by Equation (5).

$$\mathrm{Y}=93.1+2.036\mathrm{B}-176.4\mathrm{C}-0.00023{\mathrm{B}}^2+69.04{\mathrm{C}}^2-0.847\mathrm{BC}$$ (5)

TABLE 5.

ANOVA for the BBD model of CV adsorption onto CDAC.

Source	Sum of squares	Degree of freedom	Mean square	F‐value	p	Statistical significance
Model	62,344.86	9	6927.21	46.61	0.0003	Significant
A—contact time (min)	68.62	1	68.62	0.4617	0.5270	Not significant
B—CV concentration (mg L−1)	17,331.78	1	17,331.78	116.62	0.0001	Significant
C—CDAC dose (g L−1)	35,213.40	1	35,213.40	236.93	< 0.0001	Significant
AB	10.86	1	10.86	0.0731	0.7977	Not significant
AC	18.36	1	18.36	0.1235	0.7395	Not significant
BC	4032.57	1	4032.57	27.13	0.0034	Significant
A2	24.77	1	24.77	0.1666	0.7000	Not significant
B2	2.28	1	2.28	0.0154	0.9062	Not significant
C2	5512.19	1	5512.19	37.09	0.0017	Significant
Residual	743.11	5	148.62
Lack of fit	743.11	3	247.70
Pure error	0.0000	2	0.0000
Cor total	63,087.97	14

Figure 7 reveals the interaction of contact time (A), CV concentration (B), and CDAC dose (C) to control the adsorption capacity of CV on CDAC. The 2D graphs show that each factor influences the response in a modulated manner; however, it is primarily the interactions between factors that reveal the true dynamics of the system. For example, simultaneously increasing the contact time and CV concentration led to a progressive increase in the adsorption capacity, indicating a system that is kinetically and thermodynamically favorable for dye retention. It is also confirmed that CDAC dose increase leads to a slight decrease in the adsorption capacity as a consequence of distributing the adsorbed mass over a larger quantity of solid. The tightly packed contour lines in certain areas also reveal zones where the response is particularly sensitive to variations in a single factor, indicating a nonlinear interaction typical of the multifactorial systems studied using BBD. The 3D representations confirmed these trends by visualizing the response surface gradients. They highlighted the conditions leading to maximum adsorption capacity values, which are generally found at high initial concentrations and low adsorbent doses with prolonged contact times. The slightly curved shapes of the surfaces reflect the existence of significant quadratic effects, justifying the use of the BBD. Consequently, the process is dominated by the combined effect of dye concentration and contact time, whereas the adsorbent dose plays a more significant role.

FIGURE 7.

BBD 2D and 3D contours for CV adsorption onto CDAC.

The desirability profile highlights that the combination of CV concentration of 150 mg L⁻¹ and CDAC dose of 259 mg g⁻¹. An overall desirability of ≥ 0.95 indicated that these conditions almost perfectly matched the target. The values predicted by the model were accompanied by a controlled uncertainty, manifested by a standard error of 8, a 95% confidence interval of 240.3–277.22 mg g⁻¹, and a 95% prediction interval of 229.90–287.63 mg g⁻¹, confirming the robustness of the optimization and the suitability of CDAC for CV adsorption.

3.5. CDAC Regeneration

Figure 8 shows that the adsorption capacity of the material gradually decreased over the cycles, from approximately 282–253 mg g⁻¹ during the initial adsorption to 250–231 mg g⁻¹ in Cycle 1, to 225–211 mg g⁻¹ in Cycle 2, to 205–191 mg g⁻¹ in Cycle 3, to 185–184 mg g⁻¹ in Cycle 4, and finally to 180–183 mg g⁻¹ in Cycle 5, for regeneration by calcination and by distilled water. This decrease results from underlying phenomena such as the progressive deactivation of active sites, irreversible surface fouling, and—in the case of calcination—partial thermal degradation of the porous structure (Miyah et al. 2023). Regeneration with distilled water limits these alterations and results in a slower loss, primarily because of incomplete desorption and residual pore clogging (Benjelloun et al. 2024).

FIGURE 8.

CDAC regeneration cycles.

3.6. Lens culinaris Seeds Germination

The experimental observations in Figure 9 compare the germination of 10 Lens culinaris seeds on CV before adsorption and CV after adsorption. From Days 0 to 7, and daily measurements of stem length, root length, untreated CV completely blocked germination (0/10) and organ elongation (stems 0 cm, roots approximately 0.5 cm). In comparison, the medium after adsorption allowed complete germination (100%) and sustained development of stems from 1 to 12 cm and roots from 2 to 9.5 cm, that is, 19 times longer than under the toxic condition alone. These results indicate that CV is highly toxic to lentil seeds, likely through the inhibition of imbibition, enzymatic and membrane disruption, or oxidative stress. The adsorption process effectively reduces the bioavailable fraction of the dye, allowing for normal growth. From an ecological and practical standpoint, adsorption appears to be an effective decontamination strategy for making water contaminated with azo/triaryl dyes usable by plants.

FIGURE 9.

Lens culinaris seeds germination test results.

4. Conclusion

This study investigated the adsorption of CV onto activated carbon derived from CD (CDAC). Kinetic modeling indicated that adsorption was primarily governed by pseudo‐second‐order kinetics, with a secondary contribution from intraparticle diffusion. Isotherm analyses indicated a heterogeneous surface and moderate multilayer adsorption, with a favorable fit to the Freundlich and Dubinin–Radushkevich models. Thermodynamic analysis confirmed a spontaneous, endothermic process involving physisorption and weak chemisorption. The chemometric combination of the Box–Behnken experimental design and ANN accurately predicted adsorption capacities. Although the ANN outperformed the Box–Behnken model in fitting accuracy, the limited dataset restricts its role to a complementary comparative tool rather than a generalized predictive model. The desirability function indicated that the optimal adsorption capacity and removal efficiency were about 258.764 mg g⁻¹ and 86%, respectively, at [CV] = 150 mg L⁻¹, a CDAC dose of 0.5 g L⁻¹, and a time of 10 min. Post‐adsorption media showed a 100% germination rate for Lens culinaris seeds. Although this laboratory‐scale study was limited to synthetic solutions, it establishes a solid foundation for future research that should focus on complex industrial wastewater matrices, continuous‐flow systems, and long‐term adsorbent regeneration.

Author Contributions

Amal Taouda: conceptualization, investigation, methodology, writing – original draft, formal analysis, writing – review and editing. Mohammed Benjelloun: conceptualization, investigation, writing – original draft, writing – review and editing, methodology, software, formal analysis. Youssef Miyah: conceptualization, investigation, writing – original draft, methodology, writing – review and editing, validation, visualization, formal analysis, supervision. Salma Ziane: writing – original draft, writing – review and editing. Salma Ssouni: writing – original draft, writing – review and editing. Smail Manyani: writing – original draft, writing – review and editing. Noureddine El Messaoudi: writing – original draft, writing – review and editing. Fatiha Mejbar: conceptualization, writing – original draft, writing – review and editing, supervision, validation, visualization. Sanae Lairini: writing – review and editing, writing – original draft.

Funding

The authors have nothing to report.

Conflicts of Interest

The authors declare no conflicts of interest.

Data Availability Statement

Data will be made available upon request.