Implementation of Time Series Forecasting Using TFT

Code implementation of time series forecasting using Temporal Fusion Transformers(TFT) as well as handling nuances in the dataset along with Hyperparameter tuning

The Retail data from Kaggle available to us is

• Weekly historical retail sales data for more than two years,
• Anonymized information about the 45 stores, indicating the type and size of the store and
• Features containing additional store, department, and regional activity data for the given dates.

Goal of implementation is to use Temporal Fusion Transformer(TFT) to perform time-series prediction for department-wide sales for 45 stores.

Read on popular Transformer-based time series forecasting techniques here

Temporal Fusion Transformer Features

Temporal Fusion Transformer (TFT) is an attention-based deep neural network architecture enabling multi-horizon forecasting to achieve high performance while enabling new forms of interpretability, making it trustworthy.

1. Temporal Fusion Transformer (TFT) enables multi-horizon time series forecasting, which predicts a variable of interest at multiple future time steps.