Hybrid Survival Analysis Model for Predicting Automotive Component Failures

Abstract

<div class="section abstract"><div class="htmlview paragraph">A study on different survival analysis methodologies to predict when an automotive component failure can occur. By studying the various univariate and multivariate survival analysis methods and models available, we aim to develop a hybrid model that amalgamates the multiple survival analysis methods. The model takes the advantages that certain models provide and mitigate the disadvantages of other models to provide an enhanced time to failure analysis. This paper takes a deep dive into four different survival analysis models, namely, Kaplan–Meier, Cox proportional hazards model, and two ensemble models, random survival forest and gradient boosting. The novel hybrid model proposed in this paper combines the stand-alone models in a weighted sum approach to provide the best predictive capabilities. The proposed hybrid model provides a significant improvement over stand-alone models in forecasting the number of failures. The paper studies two different sets of data, which gives a detailed understanding of the effects that different models have on the data. The aforementioned techniques are employed to assess component failures in automotive vehicles, contributing to enhanced product reliability and overall user satisfaction.</div></div>