Weather Induced Flight Delay

Overview

This project aims to predict flight delays by combining machine learning and natural language processing techniques. It uses a Streamlit app for user input, processes flight route data, analyzes weather conditions at each route point, and utilizes news articles to estimate delays.

Steps

• 1. Streamlit App Interface
  • Built a Streamlit app to collect user inputs: source, destination, date, and time of travel.
• 2. ICAO Code Conversion
  • Converted the source and destination airports to ICAO format using a CSV file containing IATA and ICAO codes.
• 3. Flight Plan Retrieval
  • Used the FlightPlanDatabase API to obtain flight plans for the given source, destination, and datetime combination.
• 4. Route Point Extraction
  • Extracted all route points using the plan id from FlightPlanDatabase API.
• 5. Weather Data Collection
  • At each route point, gathered weather data using the Visual Crossing API.
• 6. Delay Calculation via ML Model
  • Calculated potential delays at each route point based on weather data using a pre-trained machine learning model.
  • Adjusted the delay estimate by dividing it by the number of route points.
  • Summed up the delay from all points and took half of this sum as part of the total delay estimate.
• 7. NLP Pipeline for Delay Prediction
  • Queried Google News API to gather news articles about flight delays specific to the source-destination pair and relevant dates.
  • Filtered news articles to those relevant to the travel date.
  • Processed these articles through a Mistral model hosted on GCP using ngrok to extract delay information.
  • Calculated the mean delay value from all articles.
• 8. Total Delay Estimation
  • The total delay estimate was calculated as 90% of the delay predicted by the ML model and 10% of the delay extracted from the NLP pipeline.

ML Model Training

• Used a dataset containing flight details and delay variables.
• Selected a sample of 10,000 rows, including 1,000 rows with weather_delay greater than 0.
• Extracted latitude and longitude for each row.
• Collected weather data for these points using the Visual Crossing API.
• Trained an XGBoost model on this data to predict delays.

Technologies Used

• Streamlit
• FlightPlanDatabase API
• Visual Crossing API
• Google News API
• XGBoost
• Mistral AI model
• NGrok