Optimal probability distribution selection analysis for airplane components MTBR data

Abstract

This thesis investigates optimal probability distribution fitting and predictive modeling for Mean Time Between Removals (MTBR) of Boeing 757 components. Historical data from 2012 are analyzed using classical distributions, including Poisson, Weibull, and Log-Normal. The Log-Normal distribution provides the best historical fit based on KS test and AIC. However, its predictive accuracy declines for 2013–2014. To address this, non-classical models such as Random Forest Regressor (RFR), Support Vector Regression, and Gaussian Mixture Models are evaluated using Mean Squared Error (MSE). Among them, RFR achieves the lowest MSE and performs consistently across B757 subtypes. The study concludes that RFR is the most reliable model for predictive maintenance. By integrating statistical fitting with machine learning, this research supports more cost-effective and data-driven maintenance planning.