Stock Market Multivariate Forecasting API
An end-to-end stock price forecasting system built using deep learning and deployed via a FastAPI backend. The project utilizes a GRU-based neural network to predict future Open and Close prices from historical market data. It incorporates data preprocessing techniques such as missing value imputation, log-return feature engineering, and MinMax scaling, along with a custom Huber directional loss to improve prediction accuracy and trend direction. The model achieved an R² score of 0.92 for Open price prediction and a directional accuracy of 56.8%, demonstrating strong performance in both value estimation and trend prediction. The system supports recursive multi-step forecasting and exposes predictions through a REST API endpoint for real-time access.
