Estimating rate of no-show passengers: A research on an airline company in Turkey

This research aims to investigate the model that best predicts no-show rates for a flight in the civil aviation industry. Predicting no-show numbers in flights is crucial for effective operational planning and revenue management for airlines. It enables airlines to manage flight capacities more efficiently by minimizing empty seats and maximizing revenue. Accurate predictions aid in optimizing pricing strategies, allowing airlines to capitalize on empty seats effectively. Features were generated by aggregating from the ticket data, including region, scheduled day part, season, day of the week (Monday, Tuesday, etc.), first half of the year (first six months) or second half (last six months), month, proximity of official holidays based on origin and destination, flight time, and average no-show rates on previous flights. The study has employed seven predictive models: Random Forest, Bagging, Ada Boost, Gradient Boosting, Extreme Gradient Boost, Neural Networks, and Decision Trees. For comparison, MAE (Mean absolute error), MSE (Mean squared error), Root Mean square error (Mean square error), R-square error (R2) and Max error values have been calculated and compared. Among the models, Decision Tree received the lowest performance score, while Gradient Boosting received the highest. Unlike previous studies, it has been observed that the previous one-month no-show averages are the most powerful variable in predicting the no-show rates of future flights. These findings were discussed in the light of previous research. Keywords: No-show, Airline Industry, Civil Aviation, Data Analytics, Regression