Modern Anomaly Detection Methods in Industry: A Comparative Analysis of Machine Learning Algorithms and Their Application to Improve the Efficiency of Manufacturing Processes

BACKGROUND The growing complexity of industrial systems and the large volume of operational data have increased the demand for automated anomaly detection to ensure production efficiency and stability. Machine learning methods provide promising solutions for identifying process deviations, but their comparative effectiveness in manufacturing environments remains insufficiently evaluated. MATERIALS AND METHODS A comparative analysis of multiple machine learning algorithms was conducted for anomaly detection in industrial time-series systems. The methods included unsupervised models (Autoencoder, Isolation Forest), supervised classifiers (Random Forest, Support Vector Machine), and density-based approaches (Local Outlier Factor), with a focus on evaluating their performance in high-dimensional, noisy environments typical of industrial datasets. RESULTS Unsupervised algorithms, particularly Autoencoder (87–89% accuracy) and Isolation Forest (84–86%), showed strong performance in environments without labeled data, making them suitable for real-world applications. Supervised classifiers achieved the highest accuracy (Random Forest: 89–91%; SVM: 88–90%) but were constrained by the availability of labeled datasets. Density- and clustering-based methods performed moderately (70–80%) in low-dimensional data but lost effectiveness as feature dimensionality increased. Implementation of machine learning-based monitoring systems demonstrated potential reductions in downtime (up to 29%), repair costs (20%), and significant improvement in productivity. CONCLUSION Machine learning methods are effective tools for anomaly detection in manufacturing systems. Algorithm selection should be tailored to the availability of labelled data, system complexity, and processing constraints. Adaptive anomaly detection frameworks provide practical benefits for industrial process control, supporting efficiency, cost reduction, and operational stability. The research reduces operational costs while simultaneously enhancing real-time monitoring, thereby improving decision-making and overall system reliability in industrial environments.