Work

An automata-machine learning integrated framework for intelligent mental health diagnostics

Year: 2026

Type: article

Abstract: Depression among the students has critical impact on student’s life and it needs to be predicted at an early stage. The unstructured nature of self-emotions are going to undetected and emerged as a critical mental health concern. Integration of machine learning (ML) with behavioural modelling address the concern and connects with student emotional state and helps to predict the depression among students at early stage. An automata-machine learning integrated approaches including ensemble of lazy and eager classifiers, behavioural modelling, minimization of finite automata are evaluated and compared in three scenarios where in the first case predictive performance on original dataset is evaluated. The second case in sighting the real life scenario student specific emotions are extracted with automata-based behavioural states to generate meaningful patterns and feature vectors for each student and these automata generated features are statistically tested and signified then embed with original dataset. Further the ML feature selection Gini index applied on automata integrated dataset to select the prominent features to enhance the predictive performance. In the third case automata based feature merging is applied which maintain the credibility of each feature and merged systematically with related features and preserve the equivalence and reduce the feature redundancy. Unlike the traditional ML based feature selection, Automata based feature merging retains the interpretability and behavioural traceability of all features. Lastly, eager learning (random forest) and lazy Learning (kstar) are ensemble and deployed to take advantage of the automata based optimized feature sets. The automata assisted scenario performs well and improves the accuracy with 14.86% where the baseline accuracy is 82.02% evaluated with random forest classifier and the accuracy with automata based behavioural modelling is recorded 96.88% and shows proposed framework is an interpretable, dynamic solution for mental health interventions in educational settings. The main contributions of the proposed framework is to extract automata based features that behaviourally connects to student emotions and mental state. The minimization of finite automata reduces the redundant factors and maintained the credibility of each feature. Further multi-metric statistical validation determines the applicability of automata extracted features.

Source: Scientific Reports

DOI: 10.1038/s41598-026-74877-7