Machine Learning Architectures for Sustainable Consumer Behaviour Modelling: Synergizing Behavioural Economics and Computational Intelligence

Sustainable technology adoption, especially when it comes to electric vehicles (EVs), depends heavily on behavioral and social factors that most traditional prediction models tend to overlook. This paper introduces a Behavior-Aware Reinforcement Learning (BARL) framework designed to bring together insights from behavioral economics, graph-based modeling of social influence, and causal inference techniques within a single learning system. The core idea treats EV adoption as a Markov Decision Process. The reward function here accounts for environmental benefits, peer effects captured through graph neural networks, and adjustments for common cognitive biases that shape real-world choices. A dedicated causal component helps the model separate genuine cause-and-effect relationships from mere correlations, making the predictions more reliable. Using the Indian Electric Vehicle Dataset for evaluation, the BARL approach was tested against several standard methods such as ARIMA, LSTM, and XGBoost. The results indicate clear improvements, with BARL lowering the RMSE by roughly 23% compared to the strongest baseline. Ablation studies further demonstrate the individual value added by the behavioral, social, and causal elements. Overall, the work underscores how human-centered modeling can offer deeper insights into EV adoption patterns and deliver a practical, scalable tool to support evidence-based policies for sustainability.