Team
Ragul Savithri Pandiyan, Kirthika Devi Periasamy, Navkaranjit Singh Randhawa, Tarandeep Singh Aulakh
Categories
Artificial Intelligence, Data Analysis, HTML, Machine Learning, Python, React Native, Time Series, Web App
Overview
A web-based forecasting platform that empowers users to upload any time-series CSV and generate accurate short- and medium-term predictions. Built with a React frontend for intuitive data selection and visualization, and a Flask backend that orchestrates three specialized deep-learning models - an LSTM for 30-step forecasts, a Patch-based Transformer (PatchTFT) for 10-step forecasts, and a TiDE encoder-decoder for 20-step forecasts. The app delivers actionable insights with metrics like RMSE, MAE, directional accuracy, and beginning-to-end accuracy. Data is preprocessed via outlier clipping and scaling before sliding-window generation, then passed through the chosen model to produce future values. Interactive charts and tables allow users to compare actual vs. predicted values and project future trends, all hosted on a scalable infrastructure for real-time inference and easy deployment.