Big Data Analytics for Predictive System Maintenance Using Machine Learning Models

Abstract

Machines are an indispensable part of every economy, playing vital roles in many sectors including production. Companies strive to produce quality products and services in order to satisfy customers and stay afloat. However, system failure leading to unprecedented downtime often impedes the delivery of goods and services, and affects businesses adversely. Consequently, the striving to keep system downtime at an ‘acceptable low’ level and mitigate associated costs is always on the rise. In this paper, Multilayer Perceptron, Support Vector Machine (SVM), and K-Nearest Neighbors (KNN) Classifier algorithms have been trained using labeled time series-data collected on production machinery to predict production machine failure within a horizon of one day and provide insight that supports the decision process for machine maintenance. By testing our models on the validation dataset, the Multilayer Perceptron neural network reliably outperformed the other models with an accuracy score of 99.99%.