CLASSIFICATION OF RECOMMENDATION SYSTEM ALGORITHMS FOR INDUSTRIAL EQUIPMENT
DOI:
https://doi.org/10.31891/2307-5732-2026-365-11Keywords:
predictive maintenance, LSTM, machine learning, recommendation systems, deep learning, IoTAbstract
The article presents a systematic classification and comprehensive analysis of recommendation system algorithms used for predictive maintenance of industrial equipment in Industry 4.0 conditions. The evolution of approaches from traditional machine learning methods to modern deep learning architectures is examined. Three main categories are analyzed: collaborative filtering methods based on similarity analysis of equipment behavior; content-based approaches utilizing characteristics and historical data; and hybrid systems combining advantages of both methodologies. Particular attention is paid to recurrent neural networks, specifically LSTM and GRU architectures, which demonstrate highest efficiency in analyzing time series sensor data with long-term dependencies. Mathematical models of main algorithms are presented, including formalization of learning and prediction processes. Comparative analysis of algorithm performance is conducted based on accuracy metrics, computational complexity, and practical applicability. Results show that LSTM networks achieve 94-97% accuracy in equipment failure prediction, while classical methods provide 88-92% accuracy with significantly lower computational complexity. Integration of recommendation systems with IoT infrastructure and cloud computing for real-time monitoring is investigated. The study determines optimal application areas of different algorithmic approaches depending on industrial equipment specifics, available computational resources, and prediction accuracy requirements. Research results provide practical recommendations for selection and implementation of predictive maintenance systems in industry. The findings indicate that choice of optimal algorithm should balance between prediction accuracy and available computational resources. For critical applications where accuracy is priority, hybrid approaches or Bi-LSTM are recommended. For systems with limited resources, optimal choice is XGBoost or GRU.
Downloads
Published
Issue
Section
License
Copyright (c) 2026 АНТОН ПАКУЛА, ВОЛОДИМИР ГАРМАШ (Автор)

This work is licensed under a Creative Commons Attribution 4.0 International License.