Building a Model with AutoML in Machine Faults Detection

Technological advancements have provided numerous benefits for production activities. The primary aim of companies is to minimize costs and avoid unforeseen expenses. Machine failures may result in excessive and unexpected costs for firms. As a result, it is of paramount importance to anticipate and prevent machine breakdowns in advance. Predictive maintenance evaluations can be valuable in recognizing equipment and machine failures. The objective of this study is to utilize automated machine learning (AutoML) in order to achieve optimal outcomes via the adjustment of parameters on machine learning (ML) algorithms aimed at detecting machine faults. The dataset comprises 77,004 rows and 13 features extracted from Google data center operators. Five different classes of machine operating status are identified, namely: evict, fail, finish, kill and lost. Among these, fail, finish and evict classes are deemed failure states. AutoML can be accessed through various libraries. This study specifically uses the Auto-Keras, Auto-Sklearn and Lazy predict libraries, all of which are open-source Python libraries. In this study, hyperparameter optimization was conducted on the random forest (RF) algorithm, which was the third best algorithm obtained through the Lazy Predict library. The code was written using Anaconda. It is important to note that while the results of these libraries are valuable, manual modifications may be necessary for optimal outcomes. Notably, the Decision Tree Classifier algorithm achieved an accuracy of 97%, which was the highest among the evaluated options.