Strategies for overcoming data scarcity, imbalance, and feature selection challenges in machine learning models for...

Area of Science:

• Industrial Engineering
• Data Science
• Machine Learning

Background:

• Predictive maintenance (PdM) uses statistical analysis to prevent equipment failures, reducing costs.
• Machine learning (ML) analyzes historical data to identify patterns and predict failures.
• Challenges in PdM include data scarcity, imbalance, and temporal dependencies.

Purpose of the Study:

• To propose an ML-based approach for predictive maintenance that addresses common data challenges.
• To enhance the accuracy and reliability of predicting equipment and system failures.

Main Methods:

• Utilized Generative Adversarial Networks (GANs) for synthetic data generation.
• Employed Long Short-Term Memory (LSTM) layers for temporal feature extraction.
• Developed failure horizons to account for temporal data characteristics.

Main Results:

• ML algorithms were trained on synthetically generated data.
• Achieved high accuracy rates: ANN (88.98%), Random Forest (74.15%), Decision Tree (73.82%), KNN (74.02%), and XGBoost (73.93%).

Conclusions:

• The proposed ML approach effectively overcomes data scarcity and imbalance in PdM.
• The method demonstrates significant improvements in predicting equipment failures, enhancing industrial maintenance strategies.