Evaluation of Preprocessing Techniques on the Performance of Artificial Intelligence Models for Gas Treatment Plant Maintenance Alert System

Abstract

Traditionally, gas treatment plants (GTP) relied on reactive and preventive maintenance strategies to manage equipment health which often leads to significant downtime and high repair costs resulting in economic loss. In order to enhance the efficiency and reliability of the machine, this study focused on an intelligent data-driven approach to predict the maintenance status. The study involves deploying ML algorithms such as RF, DT, SVM and KNN and training them with outlier datasets and various scaling techniques, robust scaling and z-score normalization. RF, DT and KNN performed well with the outliers and robust scaling with an accuracy of 100%, 100% and 86% respectively compared to SVM with an accuracy of 47%. With the removal of outliers from the dataset, all the models, RF, DT, SVM, and KNN, performed excellently well with an accuracy of 98%, 98%, 98% and 95% respectively and recall of 1.0, 1.0, 0.94 and 0.94 respectively. This shows that either of the models, especially RF and DT can be employed to develop an artificial intelligent system for real-world scenarios.