Data Science & AI · Machine Learning

Predicting Airline Passenger Satisfaction with Machine Learning

Which service factors actually predict whether a passenger is satisfied? 103,904 survey records gave an answer.

Fig. 01

01 / 01

Overview

This project applied the CRISP-DM framework to an airline satisfaction survey to predict outcomes and identify the drivers behind them.

Four models — Decision Tree, Logistic Regression, Random Forest and Gradient Boosted Trees — were tuned in RapidMiner with 5-fold cross-validation and AUC-based selection.

Gradient Boosted Trees was selected as the final model; online boarding, type of travel and inflight wifi were the strongest drivers, which pointed to where the airline should invest first.

Highlights

103,904 training and 25,976 test records across 22 predictors
Best accuracy of 96.48% (Random Forest); Gradient Boosted Trees chosen on AUC
Top 3 drivers: online boarding, type of travel, inflight wifi