The Practicality of Machine Learning for Airline Forward Sales Forecast

Abstract

Forecasting is used in the aviation industry, among other things, to estimate the number of passengers for each trip. This is advantageous for the practice of revenue management because the forecast serves as the foundation for determining the cost of each flight. In order to maximize profit, it is therefore important to know whether these forecasts are accurate. Today, the price of an airline ticket can vary significantly and quickly during the same journey, even for neighboring seats in the same cabin. Customers search for the greatest offer while airlines seek to maintain a high overall revenue and maximize their profit. Airlines use a range of computational techniques, including as pricing discrimination and demand forecasting, to increase their revenue. Demand forecasts have a crucial role in how well airline pricing and revenue management systems perform. The demand for different fare classes is assumed independent by traditional airline forecasting models, which ignore the demand's sensitivity to the airline's inventory control procedures, price variations, and the range of available alternatives. This study looks at several forecasting algorithm implementations in the context of sales booking prediction for the aviation sector, along with its benefits and drawbacks. The research is accompanied with use cases in which Time Series forecasting algorithms ARIMA and Random Forest are tested against a passenger data for American Airlines against performance indicators: confidence interval, usability, and performance. Academics, business intelligence and revenue analysts, airline operators, the revenue management team, and functional silos across the organization are the main beneficiaries of this article as sales forecast will be utilized to plan and budget flights (capacity) and to make strategic decision.