Comparing a Machine Learning Predictive Model with Federal Transit Administration (FTA)’s Default Useful Life Benchmark to Predict Replacement Costs for Revenue Vehicles

Abstract

A predictive model is developed that uses a machine learning algorithm to predict the service life of transit vehicles and calculates backlog and yearly replacement costs to achieve and maintain transit vehicles in a state of good repair. The model is applied to data from the State of Oklahoma. The vehicle service lives predicted by the machine learning predictive model (MLPM) are compared with the default useful life benchmark (ULB) of the U.S. Federal Transit Administration (FTA). The model shows that the service life predicted by the MLPM provides relatively more realistic predictions of replacement costs of revenue vehicles than the predictions generated using FTA’s default ULB. The MLPM will help Oklahoma’s transit agencies facilitate the state of good repair analysis of their transit vehicles and guide decision makers when investing in rehabilitation and replacement needs. The paper demonstrates that it is advantageous to use a MLPM to predict the service life of revenue vehicles in place of the FTA’s default ULB.