RBF Neural Network for S&P 500 Stock Closing Price Prediction

The project focuses on designing and implementing a Radial Basis Function (RBF) Neural Network for the prediction of the daily closing price of an S&P 500 stock (specifically, Apple Inc. or AAPL). Objective and Methodology: The work addresses a supervised regression problem using five input variables (Open, High, Low, Close, and Volume) to predict the next day's closing price. The RBF Network was chosen for its capacity to model non-linear relationships, utilizing a two-phase training process: non-supervised (K-Means to determine 50 centers) and supervised (weight adjustment using the Adam optimizer). Results and Conclusion: The convergence analysis showed severe overfitting (the validation error was nearly 8 times greater than the final training error). The final test set evaluation confirmed poor performance, with a Root Mean Squared Error (RMSE) of 96.82 USD, a Mean Absolute Percentage Error (MAPE) of 59.64%, and a coefficient of determination (R²) of -92.84, indicating that the model does not explain the price variance.