Predictive Maintenance on Industrial Data Using Soft Voted Ensemble Classifiers

Abstract

The IoT sector leads improvements in the Industry 4.0 revolution. Failure-prone machinery puts operations and production costs at risk. A sudden failure results in high downtime expenses and a drop in output. Predictive maintenance is a crucial area to research in order to increase industrial productivity. The purpose of the paper is to present the proper method for implementing predictive maintenance using the soft voting undersampling approach. We also highlight the impact of ensemble learning on an imbalanced dataset that is crucial problem for modelling the failure-prone machinery dataset. This paper suggests a method for enhancing predictive learning by selecting six different machine learning algorithms, including decision tree, random forest, Gradient Boosting Machines (GBM), XGBoost, Light GBM, and CatBoost classifiers. The soft voted model is proposed to enhance the performance of machine learning classifiers, which have produced better results in terms of the Fowlkes-Mallows Index (FMI) and Cohen's Kappa score. The results of our study are close to the results of predictive maintenance studies in the literature. © 2023, The Author(s), under exclusive license to Springer Nature Switzerland AG.