MICRO-DECISION IMPACT ANALYSIS USING USER CLICK STREAM DATA

Abstract

The rapid growth of e-commerce and web-based applications has increased the importance of understanding detailed user behavior during online interactions. Every action performed by a user on a digital platform, such as clicking, scrolling, hovering, or pausing on a page, represents a microdecision that contributes to the overall outcome of a browsing session. Traditional web analytics systems typically focus on high-level metrics such as page views, bounce rates, and session duration, which often fail to explain the underlying behavioral patterns that influence user conversion. This research presents a system for Micro-Decision Impact Analysis using User Click Stream Data to capture and analyze fine-grained interaction events generated during user sessions. The proposed system integrates a lightweight JavaScript tracking mechanism embedded in the client application to record interaction signals including clicks, scroll behavior, hover duration, zoom interactions, and dwell time. These events are transmitted to a FastAPI-based backend server and stored in a structured MySQL database. After session completion, feature engineering techniques are applied to generate session-level behavioral metrics such as hesitation score, scroll velocity, hover duration, zoom interaction frequency, and review reading time. A machine learning model based on the XGBoost algorithm is then trained using both publicly available datasets and real-time session data to predict user conversion probability. To improve transparency and interpretability, SHAP (SHapley Additive exPlanations) is used to quantify the contribution of each feature to the prediction outcome. The system also supports automatic model retraining when new session data accumulates and provides role-based dashboards and an analytics chatbot to deliver actionable insights for improving user experience and optimizing conversion strategies.