An Explainable Hybrid Learning Model for Stock Price Prediction: Integrating Prophet and Linear Regression in a Web-Based Investment Decision Support System

Abstract

Stock price prediction remains a challenging problem due to the volatile and nonlinear nature of financial markets. This study developed an explainable hybrid learning model that combined Prophet and Linear Regression, integrated into a web-based investment decision support system. The objectives included designing the hybrid model with technical indicators, evaluating its performance using standard metrics (MAE, MSE, RMSE, and R²), integrating the model into a user-friendly interface, and assessing usability. A mixed-method approach was adopted, incorporating quantitative techniques (for model development and performance evaluation) and qualitative usability testing. Historical data from five major tech companies (AAPL, MSFT, TSLA, AMZN, and GOOGL) were used alongside indicators like RSI, Daily Return, Volatility, and Bollinger Bands. SHapley Additive exPlanations (SHAP) was employed to interpret the influence of features on predictions. Results showed that the weighted hybrid of Prophet and Linear Regression models outperformed individual models such as Long Short-Term Memory (LSTM), Support Vector Machine (SVM), Prophet, and Linear Regression achieving lower error rates (e.g., MAE of 4.99 and R² of 0.83 for AAPL) and improved reliability across multiple stocks. Usability feedback confirmed high user satisfaction, clarity, and trust in the system's predictions and explanations. This study successfully bridged prediction performance and explainability to support better investment decisions.