An Explainable Multi-Output Learning Framework for Biosensor-Based User Interaction Analysis

Abstract

The growing emphasis on user-centered product design has increased the demand for objective and automated approaches to evaluate user satisfaction. Conventional assessment methods, including manual surveys and subjective feedback, are often time-consuming, prone to bias, and incapable of providing realtime insights into user experience. To address these limitations, this study presents a Machine Learning (ML)-based framework for predicting user satisfaction using biosensor data. Baseline models, including Decision Tree (DT), Extra Trees (ET), Linear, and Gradient Boosting (GB), are evaluated within the Classification and Regression Trees (CART) framework. To improve predictive performance, the proposed approach integrates Adaptive Boosting (AB) with CART, enhancing both classification and regression capabilities. The framework simultaneously performs interaction duration prediction as a regression task and user satisfaction classification into High and Medium categories. Experimental evaluation demonstrates that the proposed AB-enhanced CART framework achieves superior predictive performance by improving classification accuracy while reducing regression error compared with baseline models. The developed framework provides a scalable and interpretable solution for real-time user satisfaction assessment, enabling designers, manufacturers, and product developers to analyse user interactions more effectively and support the development of user-centric products and interactive systems.