Airline Passenger Satisfaction in the Digital Era: An Analysis Using Random Forest
by HAVADER Editör Ekibi
The answer you give to "were you satisfied overall?" on a post-flight survey is actually built on dozens of small experiences: how wide the seat was, how fast check-in went, how the cabin crew treated you. This study uses big data to tease apart those dozens of factors and find out which ones truly matter.
Starting from the idea that the factors shaping airline passenger satisfaction are complex and often too intertwined to be captured by simple linear relationships, the study's goal was to predict satisfaction using Random Forest on an open-access dataset of 129,880 passenger observations. This method combines many small decision trees to detect complex interactions no single simple formula could capture.
The model showed very high discriminative performance — accuracy of 0.9585 and ROC-AUC of 0.9936 — consistently outperforming a classical logistic regression model. But the genuinely interesting finding was which factors came out on top: seat comfort and online boarding emerged as the strongest predictors, while demographic traits like age and gender, and operational indicators, carried comparatively little weight.
What this study contributes is a clear priority ranking for where airlines should direct their customer-experience investments. An everyday analogy: it's similar to a hotel realizing, in its push to improve guest satisfaction, that it should focus not on lobby decor but on mattress comfort and check-in speed — prioritizing the core elements that actually shape the experience over flashy but ineffective investments.
In the end, this research combines traditional survey analysis with machine learning to show that airline passenger satisfaction isn't shaped by simple linear patterns, but by complex interactions among experiential service factors — making the practical case for data-driven decision-making in airline management.
← Blog
Starting from the idea that the factors shaping airline passenger satisfaction are complex and often too intertwined to be captured by simple linear relationships, the study's goal was to predict satisfaction using Random Forest on an open-access dataset of 129,880 passenger observations. This method combines many small decision trees to detect complex interactions no single simple formula could capture.
The model showed very high discriminative performance — accuracy of 0.9585 and ROC-AUC of 0.9936 — consistently outperforming a classical logistic regression model. But the genuinely interesting finding was which factors came out on top: seat comfort and online boarding emerged as the strongest predictors, while demographic traits like age and gender, and operational indicators, carried comparatively little weight.
What this study contributes is a clear priority ranking for where airlines should direct their customer-experience investments. An everyday analogy: it's similar to a hotel realizing, in its push to improve guest satisfaction, that it should focus not on lobby decor but on mattress comfort and check-in speed — prioritizing the core elements that actually shape the experience over flashy but ineffective investments.
In the end, this research combines traditional survey analysis with machine learning to show that airline passenger satisfaction isn't shaped by simple linear patterns, but by complex interactions among experiential service factors — making the practical case for data-driven decision-making in airline management.