Nour-Baklouti/Forecasting-Flight_Delay

Can you predict airline delays for Tunisian aviation company, Tunisair? # Forecasting-Flight_Delay Can you predict airline delays for Tunisian aviation company, Tunisair? This project follows a complete data science workflow: data preparation → exploratory analysis → modeling → evaluation → forecasting and comparison. 1. 𝐃𝐚𝐭𝐚 𝐋𝐨𝐚𝐝𝐢𝐧𝐠 𝐚𝐧𝐝 𝐏𝐫𝐞𝐩𝐚𝐫𝐚𝐭𝐢𝐨𝐧 * Importing required libraries (Pandas, NumPy, Matplotlib, Statsmodels, Scikit-learn, etc.). * Loading the flight delay dataset. * Cleaning the data and converting date columns to datetime format. * Sorting the data chronologically to preserve the time order. 2. 𝐃𝐚𝐭𝐚 𝐀𝐠𝐠𝐫𝐞𝐠𝐚𝐭𝐢𝐨𝐧 𝐚𝐧𝐝 𝐓𝐫𝐚𝐧𝐬𝐟𝐨𝐫𝐦𝐚𝐭𝐢𝐨𝐧 * Aggregating the data by day to build a time series. * Computing indicators such as daily average flight delay. * Checking the consistency and size of the time series. 3. 𝐄𝐱𝐩𝐥𝐨𝐫𝐚𝐭𝐨𝐫𝐲 𝐃𝐚𝐭𝐚 𝐀𝐧𝐚𝐥𝐲𝐬𝐢𝐬 (𝐄𝐃𝐀) * Visualizing the evolution of average delays over time. * Analyzing the distribution of delays. * Identifying trends and seasonal patterns (especially weekly seasonality). 4. 𝐃𝐚𝐭𝐚 𝐏𝐫𝐞𝐩𝐚𝐫𝐚𝐭𝐢𝐨𝐧 𝐟𝐨𝐫 𝐌𝐨𝐝𝐞𝐥𝐢𝐧𝐠 * Splitting the dataset into training and testing sets. * Scaling/normalizing the data for models that require it (e.g., LSTM). 5. 𝐌𝐨𝐝𝐞𝐥𝐢𝐧𝐠 𝐚𝐧𝐝 𝐅𝐨𝐫𝐞𝐜𝐚𝐬𝐭𝐢𝐧𝐠 Several forecasting models were implemented and compared: * ARIMA * Baseline time series model. * ACF/PACF analysis to select parameters. * SARIMA * Extension of ARIMA including weekly seasonality (7 days). * LSTM (Long Short-Term Memory) * Prophet * XGBoost 6. 𝐌𝐨𝐝𝐞𝐥 𝐄𝐯𝐚𝐥𝐮𝐚𝐭𝐢𝐨𝐧 * Comparing model performance using error metrics (e.g., RMSE, MAE). * Visual comparison of actual vs predicted values. 7. 𝐅𝐢𝐧𝐚𝐥 𝐅𝐨𝐫𝐞𝐜𝐚𝐬𝐭𝐢𝐧𝐠 Using the trained models to forecast flight delays for the year 2019. 8. 𝐂𝐨𝐦𝐩𝐚𝐫𝐢𝐬𝐨𝐧 𝐚𝐧𝐝 𝐂𝐨𝐧𝐜𝐥𝐮𝐬𝐢𝐨𝐧 * Global comparison of all tested models. * Identification of the most accurate and suitable forecasting approach.