🌌 Panel Time Series Forecasting App

A web application that showcases Panel Time Series Forecasting using an advanced Transformer model—an enhanced version of the Temporal Fusion Transformer (TFT). This model integrates multi-scale series decomposition, segment-wise attention, cross-entity attention, and an adaptive weighting mechanism for improved forecasting performance. Select from 11 diverse datasets, forecast future values, and visualize results for each entity with 7 quantiles forecasted 30 timesteps ahead, displayed in animated and interactive plots.
Built with React (Vite) and powered by Hugging Face APIs for model inference.

---

🚀 Features

• Select Datasets: Choose from 11 preloaded panel time series datasets across domains like economics, healthcare, and finance.
• One-Click Forecast: Hit the Forecast button to trigger deep learning predictions for all entities in the dataset.
• Quantile Forecasts: Visualize 7 quantile predictions per entity, plotted separately, forecasting 30 timesteps into the future.
• Interactive Visualizations: Animations and clear charts help explore model outputs dynamically.
• Dataset Details: View metadata and summary information for the selected dataset.

---

🛠️ Tech Stack

• Frontend: Vite + React + Plotly.js
• Backend API: Hugging Face Inference API
• Styling: CSS

---

📊 How It Works

1. Select a Dataset from the dropdown.
2. Click Forecast to generate predictions.
3. View:
   • Dataset details.(Total Entities,Total Datapoints,Frequency,Panel Type)
   • Forecast plots for each entity.
   • Animated quantile forecasts over time.

---

📸 Screenshots

---

📦 Setup Instructions

1. Clone the Repository:

   git clone https://github.com/Shabthana123/deep-panel-webapp.git
   cd deep-panel-webapp/frontend

2. Install Dependencies:

   npm install

3. Start the App:

   npm run dev

---

🙌 Acknowledgments

• Hugging Face for model APIs
• Plotly.js for visualization tools
• Vite + React for the frontend framework

---