Machine Failure Prediction for Predictive Maintenance Using Support Vector Machine and Isolation Forest

Abstract

Predictive maintenance has become an important means of increasing the reliability of machines and reducing unexpected failures in industrial systems. This paper proposes a machine learning-based framework for machine failure prediction using operational data. The AI4I 2020 Predictive Maintenance Dataset is used, which contains 10,000 samples with a binary target variable indicating machine failure. The dataset was preprocessed by dropping irrelevant attributes, encoding categorical variables, and standardizing numerical features. Two approaches were investigated: Support Vector Machine (SVM) as a supervised classification model and Isolation Forest as an unsupervised anomaly detection model. The performance of both methods was evaluated using accuracy, precision, recall, F1score, and confusion matrix analysis. Experimental results show that the SVM model achieves 99.45% accuracy with 96% recall for the failure class, while Isolation Forest achieves 97.25% accuracy but only 49% recall for failures, missing over half of the actual failure cases. These findings demonstrate that supervised classification with SVM is significantly more effective than unsupervised anomaly detection for failure prediction in imbalanced predictive maintenance datasets, supporting its adoption in intelligent maintenance systems aimed at reducing downtime and improving operational efficiency.