TARDIS

Project Overview

TARDIS analyzes train trajectory data to predict delays for specific routes and dates. This application processes SNCF's historical delay data through a robust cleaning pipeline, trains multiple machine learning models, and delivers predictions via an intuitive Streamlit interface.

Key Features

• Data Cleaning Pipeline: Corrects decimal values, standardizes station names, and handles missing data
• Multi-Model Approach: Compares Linear Regression, Gradient Boosting, Random Forest, and Neural Networks
• Interactive Predictions: Get delay forecasts for specific routes and dates
• Visual Analytics: Heatmaps, station comparisons, and trend visualizations

Technologies

| Component | Tools | |-----------|-------| | Frontend | Streamlit | | Backend | Python 3.9+ | | Data Processing | Pandas, NumPy | | Visualization | Matplotlib, Seaborn | | Machine Learning | Scikit-learn, TensorFlow | | Code Quality | Ruff Formatter |

Data Cleaning

The tardis_eda.ipynb notebook performs:

• Decimal Correction: Groups stations to normalize numerical values
• Text Standardization: Fixes typos in:
  • Departure station
  • Arrival station
  • Date columns
• Feature Engineering: Creates consistent temporal features
• Data Reduction: Removes non-essential columns while preserving key metrics

Frontend

1. Route Prediction Tool

   • Select departure/arrival stations
   • Choose travel date
   • Receive delay prediction

2. Historical Analysis

   • Monthly delay heatmaps
   • Station performance comparisons
   • Cancellation rate trends

3. Model Insights

   • Variable importance charts
   • Prediction confidence intervals