Table 2.

Functions of AI in self-regulation of weight management in healthy and overweight populations (n 66)

Author, year	Self-regulation use case	AI features	AI functions	Data collection instrument	Data type	Important results
Alshurafa et al., 2015	Self-monitoring (eating behaviour)	Gesture recognition	To distinguish between food types	Necklace piezoelectric sensor (vibration sensor)	Skin motion when swallowing	Successfully distinguished between liquids and solids (F-measure > 90 %), hot and cold drinks (F-measure > 90 %) and between solid food types (F-measure ∼80 %)
Amft et al., 2008	Self-monitoring (eating behaviour)	Gesture and sound recognition	To recognise dietary activities using on-body sensors	On-body sensors: (1) to detect arm movements: inertial motion sensors (three-dimensional acceleration, gyroscope and magnetometers) at the wrist and upper back (integrated into a jacket); (2) to detect chewing: ear microphone located inside the ear canal to detect bone-conducted food breakdown sounds and (3) to detect swallowing: collar sensor containing surface electromyography (EMG) electrodes and a stethoscope microphone	Arm movements, chewing cycle sounds and swallowing	Four intake gestures from arm movements and two food groups from chewing cycle sounds were detected and identified with a recall of 80–90 % and a precision of 50–64 %. The detection of individual swallows resulted in 68 % recall and 20 % precision. Sample-accurate recognition rates were 79 % for movements, 86 % for chewing and 70 % for swallowing
Amft et al., 2009	Self-monitoring (eating behaviour)	Sound recognition	To evaluate the prediction of food weight in individual bites using an ear-pad chewing sound sensor	Ear-pad chewing sound sensor	Chewing cycles and food type	Sound-based chewing recognition achieved recalls of 80 % at 60–70 % precision. Food classification of chewing sequences had an average accuracy of 94 %. Mean weight prediction error was lowest for apples (19·4 %) and the largest for lettuce (31 %)
Arif et al., 2017	Self-monitoring (physical activity)	Gesture recognition	Recognise physical activity type	Rotation forest classifier	Inertial measurement units (IMUs) placed at wrist, chest and ankles	Accurately (98 %) identified seventeen physical activities ranging from ambulation to activities of daily living
Aswani et al., 2019	Self-monitoring (eating behaviour; physical activity), optimise goal setting	Predictive analytics	To predict weight loss based on subject characteristics, step count, energy intake and counselling sessions	Bayesian classification	Secondary data	Predictive modelling framework was competitive in terms of prediction accuracy with linear SVM, logistic regression and decision tree models, which further justified the use of the utility-maximising framework and its ability to capture ‘irrational’ discounting in the decision-making of individuals participating in the intervention
Aziz et al., 2020	Self-monitoring (physical activity and energy expenditure)	Gesture recognition	To estimate energy expenditure during sitting, standing and treadmill walking using a smartwatch	LG Urbane Android smartwatch: tri-axial accelerometer, gyroscope and magnetometer	Physical movement	The activity-based models provided 7 % better energy expenditure estimation than the traditional acceleration-based models
Bastian et al., 2015	Self-monitoring (physical activity)	Gesture recognition	To discriminate between eight activity classes (lying, slouching, sitting, standing, walking, running and cycling) in a laboratory condition and walking the streets, running, cycling and taking the bus in free-living conditions	Hip-worn triaxial accelerometer	Physical movement	The performances of the laboratory-calibrated algorithm decreased for several activities when applied to free-living data. Recalibrating the algorithm with data closer to real-life conditions improved the detection of overall sitting (sensitivity: laboratory model 24·9 %; recalibrated model 95·7 %).
Bi et al., 2016	Self-monitoring (eating behaviour)	Sound recognition	To monitor and recognise food intakes in daily life	High-fidelity microphone is worn on the subject’s neck near the jaw	Acoustic signals	The accuracy of food-type recognition by AutoDietary is 84·9 %, and those to classify liquid and solid food intakes are up to 97·6 and 99·7 %, respectively.
Bouarfa et al., 2014	Self-monitoring (physical activity, energy expenditure)	Gesture recognition	Energy expenditure estimation under free-living conditions	A single ear-worn activity recognition (eAR) sensor (built-in triaxial accelerometer)	Physical movement	In free-living settings, ten different types of physical activities (i.e., lying down, standing, computer work, vacuuming, going up and downstairs, slow walking, brisk walking, slow running, fast running and cycling) were predicted
Chung et al., 2018	Self-monitoring (eating behaviour and physical activity)	Gesture recognition	To detect the patterns of temporalis muscle activities during food intake and other physical activities	A glasses-type device with an in-built EMG or piezoelectric strain sensor and attaching them directly onto human skin	Skin movement	The average F1 score of the classification among the featured activities was 91·4 %.
Dijkhuis et al., 2018	Self-monitoring (physical activity), optimise goal setting	Predictive analytics	To predict the likelihood of a user to achieve a daily personal step goal	Wrist-worn activity tracker, the Fitbit Flex	Physical activity	In 80 % of the individual cases, the random forest algorithm was the best performing algorithm
Dobbins et al., 2017	Self-monitoring (physical activity)	Gesture recognition	To distinguishing physical activity	Tri-axial accelerometers and a heart-rate monitor	Physical movement	The results showed an improvement in recognition accuracy as compared with existing studies, with accuracies of up to 99 % and sensitivities of 100 %
Dong et al., 2014	Self-monitoring (eating behaviour)	Gesture recognition	To automatically detect periods of eating (free-living condition)	iPhone 4 (accelerometer and gyroscope) placed inside a pouch, wrapped snugly around the forearm	Wrist motion (linear and rotational)	Results show an accuracy of 81 % for detecting eating at 1 s resolution
Ermes et al., 2008	Self-monitoring (physical activity)	Gesture recognition	To recognise sports performed by the subjects	3-D accelerometers on hip and wrist and GPS information	Physical movement	The total accuracy of the activity recognition using both supervised and unsupervised data was 89 % that was only 1 % unit lower than the accuracy of activity recognition using only supervised data. The accuracy decreased by 17 % unit when only supervised data were used for training and only unsupervised data for validation.
Everett et al., 2018	Self-monitoring (physical activity), self-control (physical activity), optimise goal setting	Real-time analytics with personalised micro-interventions	Automatically translate various raw mobile phone data into insights about user’s life habits. Provide personalised, contextual, just-in-time, just-in place recommendations (tailored messages)	Phone accelerometer	Physical movement	Physical activity increased by 2·8 metabolic equivalents of task (MET) – hours per week, weight reduced by 1·6 kg (P < 0·001) (˜2 %). BMI declined by 0·6 kg/m2 (P < 0·001), and waist circumference was reduced by 1·4 cm (P < 0·01)
Fontana et al., 2014	Self-monitoring (eating behaviour)	Gesture recognition	To objectively monitor ingestive behaviour in free-living	A jaw motion sensor, a hand gesture sensor and an accelerometer (integrated into a device and wirelessly interface to a smartphone)	Jaw and hand motion	The system was able to detect food intake with an average accuracy of 89·8 %
Forman et al., 2018	Self-monitoring (eating behaviour), self-control (eating behaviour), optimise goal setting	Real-time analytics with personalised micro-interventions	Predict dietary lapses and deliver a targeted intervention designed to prevent the lapse from occurring	Ensemble methods (combining weighted vote of predictions from random forest, Logit. Boost, Bagging, Random Subspace, Bayes Net)	Ecological momentary assessment (EMA) 6 times a day + ad hoc entry of lapse event	Of the twenty-one possible triggers, 29 % of intervention triggers were based on time of day, 16·7 % based on low motivation and 10 % based on fatigue. The remaining eighteen triggers were identified as risk factors < 10 % of the time. There was a reduction in unplanned lapses. Participants averaged a 3·13 % weight loss
Forman et al., 2019a	Self-monitoring (eating behaviour), self-control (eating behaviour), optimise goal setting	Real-time analytics with personalised micro-interventions	Predict dietary lapses and deliver a targeted intervention designed to prevent the lapse from occurring	Ensemble methods (e.g., combining weighted vote of predictions from Random Forest, Logit. Boost, Bagging, Random Subspace, Bayes Net)	EMA 6 times a day + ad hoc entry of lapse event	Weight Watcher (WW) + OnTrack (OT) participants reported an average of 29·72 (sd = 29·11) lapses during the 10-week study period, and the frequency decreased through time. Weight losses were greater for WW + OT (M = 4·7 %, se = 0·55) than for WW (M = 2·6 %, se = 0·80)
Forman et al., 2019b	Self-monitoring (eating behaviour), self-control (eating behaviour), optimise goal setting	Real-time analytics with personalised micro-interventions	Predict dietary lapses and deliver a targeted intervention designed to prevent the lapse from occurring	Reinforcement learning algorithm	Physical activity measured in minutes of moderate-to-vigorous physical activity (MVPA) using Fitbit Flex or a similar type of Fitbit wrist-worn activity tracker	Proposed system achieved weight losses equivalent to existing human coaching programmes (non-optimised (NO) = 4·42 %, individually optimised (IO) = 4·56 %, group-optimised (GO) = 4·39 %) at roughly one-third the cost (1·73 and 1·77 coaching hours/participant for IO and GO, v. 4·38 for NO)
Fullerton et al., 2017	Self-monitoring (physical activity)	Gesture and image recognition	To recognise activity and sub-category activity types through the use of multiple body-worn accelerometers in a free-living environment	Nine body-worn accelerometers for a day of free living	Physical movement	The recognition accuracy of 97·6 %. Controlled and free-living testing provided highly accurate recognition for sub-category activities (> 95·0 %). Decision tree classifiers and maximum features demonstrated to have the lowest computing time
Goldstein et al., 2018	Self-monitoring (eating behaviour)	Predictive analytics	Predict dietary lapses	Ensemble methods (combining weighted vote of predictions from random forest, Logit. Boost, Bagging, Random Subspace, Bayes Net)	EMA 6 times a day + ad hoc entry of lapse event	Participants responded to an average of 94·6 % of EMA prompts (range = 85·2–98·9 %) and compliance remained relatively stable throughout the study
Goldstein et al., 2020	Self-monitoring (eating behaviour), self-control (eating behaviour) optimise goal setting	Real-time analytics with personalised micro-interventions	To measure dietary lapses and relevant lapse triggers and provide personalised intervention using machine learning	Decision tree	EMA 6 times a day + ad hoc entry of lapse event	Average of 4·36 lapses per week (sd = 1·46). Participants lost an average of 2·6 % of their starting weight at mid-treatment and 3·4 % at end-of-treatment
Hegde et al., 2017	Self-monitoring (physical activity)	Gesture recognition	To propose an insole-based activity monitor – SmartStep, designed to be socially acceptable and comfortable	Insole-based sensor system: contains a 3D accelerometer, a gyroscope The wrist sensor was worn on the wrist of the dominant hand like a wristwatch The ActivPal (AP), a commercially available positional sensor module worn on the thigh as a criterion measure during the free-living study. It classifies individuals’ activities into periods spent sedentary, standing and stepping	Physical movement	The overall agreement with ActivPAL was 82·5 % (compared with 97 % for the laboratory study). The SmartStep scored the best on the perceived comfort reported at the end of the study
Hezarjaribi et al., 2018	Self-monitoring (energy intake)	Speech recognition	To facilitate nutrition monitoring using speech recognition and text mining	Smartphone microphone	Speech	Speech2Health achieves an accuracy of 92·2 % in computing energy intake
Hossain et al., 2020	Self-monitoring (eating behaviour)	Image recognition	To detect and count bites and chews automatically from meal videos	Video recorder	Video images	Mean accuracy of 85·4 % ± 6·3 % concerning manual annotation was obtained for the number of bites and 88·9 % (± 7·4 %) for the number of chews
Hua et al., 2020	Self-monitoring (physical activity)	Gesture recognition	To classify nine different upper extremity exercises	Triaxial IMU	Physical movement (kinematics)	Random forest models with flattened kinematic data as a feature had the greatest accuracy (98·6 %). Using the triaxial joint range of motion as the feature set resulted in decreased accuracy (91·9 %) with faster speeds
Huang et al., 2017	Self-monitoring (eating behaviour)	Gesture recognition	Recognise eating behaviour and food type	On-board real-time decision algorithm; chewing detection algorithm; decision trees	Electromyography embedded in wearable glasses connected to the smartphone	96 % accuracy in detecting chewing and classifying five types of food
Jain et al., 2018	Self-monitoring (physical activity)	Gesture recognition	To classify activities using built-in sensors of smartphones	The phone kept in the front pocket of the subject’s trousers, built-in accelerometer and gyroscope sensor	Physical movement	Average activity classification accuracy achieved using the proposed method was 97·12 %
Jiang et al., 2020	Self-monitoring (energy intake)	Image recognition	To develop a deep model-based food recognition and dietary assessment system to study and analyse food items from daily meal images (e.g., captured by smartphone)	Existing datasets	Images	The system was able to recognise food items accurately with top-1 accuracy of 71·7 % and top-5 accuracy of 93·1 %
Juarascio et al., 2020	Self-monitoring (eating behaviour)	Predictive analytics	To detect changes in HRV to in turn detect the risk of experiencing an emotional eating episode in an ecologically valid setting	Empatica E4 wrist-sensor (photoplethysmography: non-invasive the optical measurement that can derive cardiovascular features from light absorption of the skin) and EMA six prompts per day (participants were also instructed to self-report immediately following an emotional eating episode and answer the same questions)	Heart rate variability (HRV)	Support vector machine (SVM) models using frequency-domain features achieved the highest classification accuracy (77·99 %), sensitivity (78·75 %) and specificity (75·00 %), though were less accurate at classifying episodes (accuracy 63·48 %, sensitivity 62·68 % and specificity 70·00 %) and did not meet acceptable classification accuracy
Kang et al., 2019	Self-monitoring (physical activity, energy expenditure)	Gesture recognition	To predict the energy expenditure of physical activities	AirBeat system: built-in patch-type sensor module for wireless monitoring of heart rate, exercise index, ECG and a three-axial acceleration motion detector	Physical movement, HR, exercise index, ECG, humidity, and temperature	RMSE of 0·1893 and R 2 of 0·91 for the energy expenditures of aerobic and anaerobic exercises
Kim et al., 2015	Self-monitoring (physical activity)	Gesture and image recognition	To recognise sedentary behaviour	Two accelerometers (waist over the right hip and right thigh) and a wearable camera (around the neck using a lanyard)	Physical movement	ActivPAL showed the most accurate estimate of total sedentary time with MAPE of 4·11 % and percentage of bias of –3·52 %
Korpusik et al., 2017	Self-monitoring (energy intake)	Speech recognition	To automatically extract food concepts (nutrients and energetic intake) from a user’s spoken meal description	Amazon Mechanical Turk (AMT) where Turkers were to record ten meal descriptions	Speech	83 % semantic tagging accuracy
Kyritsis et al., 2019	Self-monitoring (eating behaviour)	Gesture recognition	To automatically detect in-meal food intake cycles using the inertial signals (acceleration and orientation velocity) from an off-the-shelf smartwatch	Off-the-shelf smartwatch (acceleration and orientation velocity)	In-meal bite detection	Achieved the highest F1 detection score (0·913 in the leave-one subject-out experiment) as compared with existing algorithms
Lin et al., 2012	Self-monitoring (physical activity, energy expenditure)	Gesture recognition	To recognise physical activities and their corresponding energy expenditure	Motion sensors and an ECG sensor	Physical movement and ECG	Recognition accuracies using decision trees in the cross-validations ranged from 95·52 to 97·70 %
Lin et al., 2019	Self-monitoring (physical activity, energy expenditure)	Image recognition	To estimate energy expenditure of physical activity in gyms	Kinect for XBOX 360 sensors (depth and motion sensing)	Physical movement (Kinect skeletal data)	The measured and predicted metabolic equivalents of task exhibited a strong positive correlation
Liu et al., 2012	Self-monitoring (physical activity, energy expenditure)	Gesture recognition	To recognise the physical activity	Two triaxial accelerometers (at hip and wrist), and one ventilation sensor secured to the abdomen (AB) at the level of the umbilicus	Ventilation (abdomen), motion (hip), motion (wrist)	Correctly recognised the thirteen activity types 88·1 % of the time, which is 12·3 % higher than using a hip accelerometer alone. Also, the method predicted energy expenditure with a root mean square error of 0·42 MET, 22·2 % lower than using a hip accelerometer alone
Liu et al., 2015	Self-monitoring (physical activity and energy expenditure), self-control (eating, physical activity), optimise goal setting	Gesture recognition and real-time analytics with personalised micro-interventions (1 subject)	To recognise physical activity and provide health feedback	Android phone with built-in accelerometer and magnetometer	Nine basic daily physical activities: walking, jogging, ascending and descending stairs, bicycling, travelling up in an elevator, travelling down in an elevator, using an escalator and remaining stationary	Achieved an average recognition accuracy of 98·0 % with a minimised energy expenditure
Liu et al., 2018	Self-monitoring (eating behaviour)	Image recognition	To recognise food items	Camera on smartphones	Food type	(1) outperformed existing work in terms of food recognition accuracy (top-1: 77·5 %; top-5: 95·2 %); (2) reduced response time that is equivalent to the minimum of the existing approaches and (3) lowered energy consumption which is close to the minimum of the state of the art
Lo et al., 2020	Self-monitoring (eating behaviour)	Image recognition	To estimate the portion size of food items consumed	Camera	Portion size (often commonly seen food categories including burger, fried rice, pizza, etc. Each category has twenty food models with different shape geometries and portion size)	Mean accuracy of up to 84·68 %
Lopez-Meyer et al., 2010	Self-monitoring (eating behaviour)	Image recognition	To describe the detection of food intake by a support vector machine classifier trained on-time history of chews and swallows	Videotaped by a camcorder to capture subject activity	Chewing and swallowing	The highest accuracy of detecting food intake (94 %) was achieved when both chews and swallows were used as predictors
Mo et al., 2012	Self-monitoring (physical activity, energy expenditure)	Gesture recognition	To estimate energy expenditure	Wireless wearable multi-sensor integrated measurement system (WIMS): two triaxial accelerometers, worn at the hip and wrist	Body motion and breathing	Under free-living conditions, WIMS correctly recognised the activity intensity level 86 % of the time
Montoye et al., 2016	Self-monitoring (physical activity)	Gesture recognition	To recognise physical activity type	Oxycon Mobile portable metabolic analyser and four accelerometer-based activity monitors	Physical movement	Overall classification accuracy for assessing activity type was 66–81 % for accelerometers mounted on the hip, wrists and thigh, which improved to 73–87 % when combining similar activities into categories. The wrist-mounted accelerometers achieved the highest accuracy for individual activities (80·9–81·1 %) and activity categories (86·6–86·7 %); accuracy was not different between wrists. The hip-mounted accelerometer had the lowest accuracy (66·2 % individual activities, 72·5 % activity categories)
Päßler et al., 2014	Self-monitoring (eating behaviour)	Sound recognition	To recognise chewing sounds	Microphones applied to the outer ear canal	Chewing sound	Precision and recall over 80 % were achieved by most of the algorithms
Parkka et al., 2010	Self-monitoring (physical activity)	Gesture recognition	To automatically recognise the physical activity	Nokia wireless motion bands 3-D accelerometer	Physical movement	Overall accuracy was 86·6 and 94·0 % after classifier personalisation
Pouladzadeh et al., 2014	Self-monitoring (energy intake)	Image recognition	To estimate food energy and nutrition	The built-in camera of smartphones or tablets	Food size, shape, colour and texture. Food portion was estimated based on the area	Accuracies in detecting single, non-mixed and mixed foods were 92·21, 85 and 35–65 %, respectively
Pouladzadeh et al., 2015	Self-monitoring (eating behaviour)	Sound recognition	To estimate food energy and nutrition using a cloud-based support vector machine (SVM) method	Built-in camera of smartphones or tablets	Food size, shape, colour and texture. Food portion was estimated based on the area	By using a cloud computing system in the classification phase and updating the database periodically, the accuracy of the recognition step has increased in single food portion, a non-mixed and mixed plate of food compared with LIBSVM
Rabbi et al., 2015	Self-monitoring (physical activity), self-control (physical activity), optimise goal setting	Real-time analytics with personalised micro-interventions	To automatically (1) track physical activity, (2) analyse activity and food logs to identify frequent and nonfrequent behaviours and (3) generate personalised suggestions that ask users to either continue, avoid or make small changes	Accelerometer and GPS, smartphone food logging	Four most common daily physical activities – walking, running, stationary (sitting or standing) and driving	Physical activity increased by 2·8 metabolic equivalents of task (MET) – hours per week (sd 6·8; P = 0·02)
Rachakonda et al., 2020	Self-monitoring (energy intake)	Image recognition	To automatically detect, classify and quantify the objects from the plate of the user	A camera attached to glasses	Food type, amount, time of eating	The iLog model has produced an overall accuracy of 98 % with an average precision of 85·8 %
Sazonov et al., 2010	Self-monitoring (eating behaviour)	Sound recognition	To detect acoustical swallowing	Throat microphone located over laryngopharynx	Swallowing sounds	Average weighted epoch recognition accuracy for intra-visit individual models was 96·8 % which resulted in 84·7 % average weighted accuracy in detection of swallowing events
Sazonov et al., 2012	Self-monitoring (eating behaviour)	Gesture recognition	To detect periods of food intake based on chewing	Piezoelectric strain gauge sensor	Jaw movement	Classification accuracy of 80·98 % and a fine time resolution of 30 s
Sazonov et al., 2016	Self-monitoring (physical activity, energy expenditure)	Gesture recognition	To describe the use of a shoe-based wearable sensor system (SmartShoe) with a mobile phone for real-time recognition of various postures/physical activities and the resulting EE	Five force-sensitive resistors (integrated into a flexible insole) and an accelerometer	Physical movement	Results showed a classification accuracy virtually identical to SVM (∼95 %) while reducing the running time and the memory requirements by a factor of > 103
Spanakis et al., 2017a	Self-monitoring (eating; emotions)	Predictive analytics	Analyse individual states of a person status (emotions, location, activity, etc.) and assess their impact on unhealthy eating	Classification decision trees; hierarchical agglomerative clustering	EMA 10 times a day + ad hoc entry of lapse event	Participants were clustered into six groups based on their eating behaviour and specific rules that discriminate which conditions lead to healthy v. unhealthy eating
Spanakis et al., 2017b	Self-monitoring (eating; emotions), self-control (eating)	Real-time analytics with personalised micro-interventions	Analyse user-specific data, highlight most discriminating patterns that lead to unhealthy eating behaviour and providing feedback (personalised warning messages before a possible unhealthy eating event)	Classification decision trees and hierarchical agglomerative clustering	EMA 6 times a day + ad hoc entry of lapse event	Participants reported on average 3·6 eating events (sd = 1·1) per day
Stein et al., 2017	Self-monitoring (eating; physical activity; emotions), self-control (eating; physical activity), optimise goal setting	Real-time analytics with personalised micro-interventions	Predict dietary lapses and provide adaptive semi-individualised feedback to users regarding their eating behaviour	Used a previously used algorithm which used decision tree	Chatbot	Percentage of healthy meals increased by 31 % of total meals logged at baseline to 67 % within 21 weeks; the percentage of unhealthy meals decreased by 54 %. Users averaged 2·4 kg or 2·4 % weight loss, and 75·7 % (53/70) of users lost weight in the programme
Tao et al., 2018	Self-monitoring (physical activity, energy expenditure)	Gesture and image recognition	To estimate energetic expenditure	Camera, two wearable accelerometers	Physical movement, HR, exercise index, humidity and temperature	The fusion of visual and inertial data reduces the estimation error by 8 and 18 % compared with the use of visual-only and inertial sensor only, respectively, and by 33 % compared with a MET-based approach
Thomaz et al., 2015	Self-monitoring (eating)	Sound recognition	Recognise eating behaviour (chewing and biting sound from ambient noises)	SVM; nearest neighbours and random forest	Wrist-worn audio recording device	Detected eating with 86·6 % accuracy
Vathsangam et al., 2011	Self-monitoring (physical activity, energy expenditure)	Gesture recognition	To estimate energy expenditure during treadmill walking	Inertial measurement unit (IMU): triple-axis accelerometer, triaxial gyroscopes	Physical movement	Combining accelerometer and gyroscope information leads to improved accuracy compared with using either sensor alone
Vathsangam et al., 2014	Self-monitoring (physical activity, energy expenditure)	Gesture recognition	To detect physical activity using different features	Phone-based triaxial accelerometer	Physical movement	Feature combinations corresponding to sedentary energy expenditure, sedentary heart rate and sex alone resulted in errors that were higher than speed-based models and nearest-neighbour models. Size-based features such as BMI, weight and height produced lower errors. Weight was the best individual descriptor followed by height.
Walker et al., 2014	Self-monitoring (eating behaviour)	Sound recognition	To automatically detect ingestion	Throat microphone located over laryngopharynx	Eating sound	> 94 % of ingestion sounds are correctly identified with false-positive rates around 9 % based on 10-fold cross-validation
Wang et al., 2019	Self-monitoring of weight loss progress	Predictive analytics	Predict weight loss based on socio temporal context	Linear regression; stochastic gradient descent (SGD)	Secondary data	Weight loss can be predicted based on temporal-social information
Yunus et al., 2019	Self-monitoring (energy intake)	Image recognition	To automatically estimate food attributes such as ingredients and nutritional value	Existing image datasets	Food type and portion	Results showed the top 1 classification rate of up to 85 %
Zhang et al., 2017	Self-monitoring (eating behaviour)	Gesture and image recognition	To detect eating	Wrist-worn sensor (Microsoft Band 2-accelerometer and gyroscope) and an HD webcam camera	Eating and non-eating gestures	Results showed a correlation between feeding gesture count and energetic intake in unstructured eating (r = 0·79, P-value = 0·007)
Zhang et al., 2018	Self-monitoring (physical activity)	Wi-Fi signal recognition	To recognise general physical activity	The software platform, a signal transmitter and a signal receiver	Wi-Fi signal	Results showed a recognition rate of the general presence of physical activity of 99·05 %, an average recognition rate of 92 % when detecting four common classes of activities
Zhou et al., 2019	Self-monitoring (physical activity)	Predictive analytics	Predict exercise lapse	SVM; logistic regression	Secondary data	Discontinuation prediction score (DiPS) makes accurate predictions on exercise goal lapse based on short-term data. The most predictive features were steps and physical activity intensity
Zhou et al., 2020	Self-monitoring (physical activity), self-control (physical activity), optimise goal setting	Real-time analytics with recommendations	Adaptively compute personalised step goals that are predicted to maximise future physical activity for each participant based on all the past steps’ data and goals of each participant	Behavioural analytics algorithm (BAA)	Phone accelerometer	Participants in the intervention group had a decrease in mean (sd) daily step count of 390 (490) steps between run-in and 10 weeks, compared with a decrease of 1350 (420) steps among control participants (n 30; P = 0·03). The net difference in daily steps between the groups was 960 steps (95 % CI 90, 1830 steps)