Airfare Forecasting: A Deep Learning Approach to Flight Price Prediction

Ahstract-This study addresses the challenging issue of flight price prediction, which is a concern to both consumers and airline companies. To tackle this problem, we employed three traditional machine learning models alongside three advanced deep learning models: Recurrent Neural Networks (RNN), Gated Recurrent Units (GRU), and Long Short-Term Memory Networks (LSTM), in both unidirectional and bidirectional forms, to predict air ticket prices. These models were selected for their proficiency in processing time-series data. For our experiments, we used our own dataset, which we collected through web scraping from Google Flights. This dataset comprises flight data over seven months for flights between six major US airports. This dataset was employed to train each model using a windowed approach followed by a date-based train-test split. We divided the dataset into 80% for training, 10% for validation, and 10% for testing based on the flight date. Our overall experimental results demonstrated that LSTM models outperform all other traditional machine learning and deep learning models utilized. Notably, the Bi-LSTM model achieved an average Mean Absolute Error (MAE) of 46.7 and a Mean Squared Error (MSE) of 3852.6. In addition to the overall evaluation, we divided the dataset into three categories: short, medium, and long, based on the distance of the flights and evaluated the traditional machine learning and deep learning models. Even when evaluated based on distance categories, LSTM models outperformed all the other models. Additionally, all the deep learning models performed better than all the traditional machine learning models in airfare prediction.