A Systematic Literature Review of Failure Prediction in Production Environment Using Machine Learning Technique

Abstract

Context: Process continuity is one of the fundamental quality attributes of a production environment. The accurate prediction of a process failure is a significant challenge for the effective management of the production delivery process.

Objective: The primary aim of this paper is to present a systematic review of studies related to the prediction of failure in production environments using machine learning techniques. Several research questions were identified and investigated in this review, with the goal of providing a comprehensive summary and analyses, and discussing various viewpoints concerning failure prediction measurements, datasets, metrics, measures of evaluation, individual models, and ensemble models.

Method: The study employed the usual systematic literature review methodology and was limited to the most widely used digital database libraries for computer science from January 2016 to May 2021.

Results: We examined 42 relevant research published in peer-reviewed journals and conference proceedings. The findings indicate that there is just a small amount of activity in the region of the production environment using failure prediction compared with other service quality attributes. SVM, RF, DT, LR, and LSTM were the most used ML techniques employed in the selected primary studies, and the most accurate is the prediction model using ANN. Most studies concentrated on regression problems and used supervised kinds of machine learning. Individual and ensemble prediction models were used in most investigations, with the number of studies using each type being nearly equal.

Conclusion: According to the findings of this comprehensive literature analysis, ensemble models outperformed individual models in terms of accuracy prediction and have been found to be helpful models in predicting faults or unexpected events. However, their use is rather infrequent, and there is a pressing need to put these and other models to use in the real world to a large number of datasets with a diverse collection of datasets in order to improve the accuracy and consistency of the findings.