Real-Time User Satisfaction Modeling Using Biosignals and Adaptive Boosting Techniques

Abstract

The rapid rise of user-centered product design has increased the need for understanding user satisfaction through objective measures. According to recent studies, over 65% of organizations report product usability as a key factor in customer retention, while 75% of users disengage after unsatisfactory interactions. Traditional manual surveys and subjective feedback mechanisms are timeconsuming and often biased, providing limited real-time insights into user experience. To address this gap, this study proposes a Machine Learning (ML)-based framework that leverages biosensor data for predicting user satisfaction. Existing methods such as Classification and Regression Trees (CART) using Decision Tree (DT), Extra Trees (ET), Linear Regression (LR), and Gradient Boosting (GB) serve as baseline models. The proposed method introduces a CART framework with Adaptive Boosting (AdaBoost), enhancing both regression and classification performance. The output comprises two dimensions: interaction duration prediction (regression) and user satisfaction classification (High and Medium categories). Experimental results demonstrate that the proposed model significantly reduces error rates and improves accuracy compared to existing methods, ensuring robust predictions. This methodology offers a scalable and intelligent solution to quantify user satisfaction in real time, supporting designers, manufacturers, and product developers in delivering more engaging and user-friendly systems.