Performance Evaluation of Artificial Intelligence-Based Ticket Demand Forecasting

Abstract

Complex policies are generally employed by corporations to fluctuate the price of the products. The airline industry is amongst the most advanced in the application of the revenue management in an effort to optimize their revenue. Air ticket market is not uniform in all countries and it depends on the supply volume and pattern as well as demand. The paper explores the issue of ticket demand prediction and customer support behavior along with real-life Customer Support Ticket Dataset of Kaggle (8,469 records with rich metadata). Once the exploratory analysis of the ticket priorities and word cloud visualization of the resolved and unresolved issues were completed, the text preprocessing with tokenizing, normalizing of cases, stopwords elimination, and TF-IDF features extraction was conducted to prepare the textual fields to be modeled. This was followed by the division of the data into training (80) and testing (20) set and the application of a Gradient Boosting model, which predicts the demand of tickets based on its high ensemble learning properties to transform a number of weak learners into a powerful predictor. The assessment of the performance according to R² RMSE MSE MAE indicated that the model has outstanding predictive power with R² value 95.0 percent and low error measures of RMSE 0.47, MSE 0.22, and MAE 0.30 being obtained and with the close clustering of the actual and predicted plots and also close-to-the-data residual values. It was demonstrated that Gradient Boosting was more effective than the traditional and deep learning baselines such as Linear Regression (72.3%), LSTM (70.0%), and KNN (65.0%). Findings support the validity of enhancing-based ensemble techniques in demand prediction in customer support settings, which is important in terms of resource allocation, workload, and service optimization.