Predictive Models with XAI: A Comparative Study of Enhancing Airline Customer Satisfaction

In today's airline industry, it is crucial to keep customer happy and satisfied. Airlines are always looking for ways to improve their services and relationships with passengers so they can make necessary improvements. The primary objective of this study is to predict customer satisfaction based on various parameters and identify areas in which the airline can enhance its services to generate more satisfied customers. The models were trained on an Airlines Customer Satisfaction dataset, provided by IIT Roorkee in 2020 containing 129,880 rows and 24 columns, including the target variable "satisfaction". The study employed two different approaches to make predictions: a Blackbox approach using a deep neural network which obtained an overall accuracy of 92% and a Glassbox approach using a decision tree which reached 94% accuracy. Both approaches were evaluated by standard measures such as accuracy, loss, precision, recall, f1-score, and confusion matrices. In addition, LIME and SHAP approach were applied to the models to retrieve further insights into the predictions and feature importance. The results indicated that XAI explains the Blackbox approach well. The Glassbox approach, as it is explainable on its own, does not require XAI. Therefore, after comparing the models' accuracy and level of explainability, researchers recommend the use of the Glassbox approaches for airline customer satisfaction.