This project implements a Temporal Fusion Transformer (TFT) to forecast Microsoft (MSFT) stock prices. It features a full pipeline, including advanced feature engineering with technical indicators and time-based covariates. The model's performance is benchmarked against DeepAR and N-BEATS for a comprehensive analysis.

This project implements a Temporal Convolutional Network (TCN) for time series forecasting. Using synthetic data (including trend, seasonality, and noise), the model’s ability to learn complex patterns and provide probabilistic forecasts is demonstrated with the Darts library.

This project implements and compares deep learning models (DeepAR & N-BEATS) for multivariate time series forecasting of daily temperature. Using the Darts library, it showcases feature engineering with past and future covariates, probabilistic forecasting, and robust backtesting on the Daily Delhi Climate dataset.

DeepAR (LSTM+Gaussian) and N-BEATS models forecast GOOG (2015–2025) with covariates, scaling, and robust splits. Trained via PyTorch Lightning (early stopping, checkpoint, LR scheduling). Grid search optimized hyperparams. Backtesting showed reliable results (MAE 0.0936, MAPE 12%).

This repository contains a time series forecasting project using the Google Play Store dataset. It systematically compares RNN, LSTM, and GRU models, optimized via Keras Tuner, to predict future app metrics and evaluate performance.

This project forecasts MSFT stock prices by comparing four advanced deep learning models: TFT, TCN, DeepAR, and N-BEATS. It uses a robust pipeline with technical indicators as features. The TCN model achieved the highest accuracy, demonstrating a comprehensive approach to time-series model selection.

This project uses a Temporal Fusion Transformer (TFT) model to predict air passenger traffic. Key steps include normalizing monthly data, engineering time-based features (year, month), and training with the Darts library to capture trends and seasonality for accurate forecasting.