Enhancing tabular data analysis for classification of airline passenger satisfaction using TabNet deep neural network

In an era of air travel, understanding and enhancing passenger satisfaction are pivotal to the success of airlines and the overall passenger experience. Analyzing airline passenger satisfaction using tabular data can pose various challenges, both when employing classical statistical methods and when leveraging machine learning and deep learning techniques. On the one hand, statistical approaches pose various challenges including limited feature engineering techniques, the assumption of linearity of the data sets and limited predictive power. Then again, the use of machine learning and deep learning techniques may face other challenges such as the problem of overfitting, difficulty in interpreting data, intensive resource requirements, and the generalization problem in deploying machine learning-based methods. This paper presents a novel deep learning approach utilizing TabNet to classify airline passenger satisfaction. Leveraging a comprehensive dataset comprising various passenger-related attributes, our TabNet-based model demonstrates exceptional performance in distinguishing between satisfied and dissatisfied passengers. Our model’s robustness in handling tabular data, underscores its power as a valuable tool for the aviation industry. Comparing out results to recent papers show that out model outperforms these studies in terms of accuracy, precision, recall and area under the curve. The results show that our TabNet Network model outperforms all implemented machine learning models by reaching respectively the following results: 96.47%, 96.41% and 96.24% for accuracy, F1-score and G-mean score.