Comparison of Online Learning Vs Batch Learning in Predictive Maintenance Systems
M Ramaraju, Kethavath Nagu, B Surekha
This study presents a comparative analysis of batch learning and online learning methodologies in
predictive maintenance (PdM) systems. The objective is to evaluate the performance, resource efficiency,
and adaptability of these two approaches in the context of equipment failure prediction. Batch learning
models, trained on comprehensive datasets, achieved an accuracy of 91.2% but were characterized
by longer prediction latencies (6.0 seconds), higher memory usage (4.0 GB), and increased CPU
consumption (70%). Conversely, online learning models, which continuously update with new data,
demonstrated a slightly lower accuracy of 88.5%, yet they excelled in real-time performance with a
latency of 2.0 seconds, reduced memory usage (1.5 GB), and lower CPU utilization (40%). Additionally,
online learning models showed greater adaptability, achieving a 90.3% adaptability rate, and required
significantly less training time (45 minutes) compared to the 12 hours needed for batch learning. This
study highlights the trade-offs between batch and online learning approaches, offering valuable insights
for optimizing predictive maintenance systems where real-time data processing and resource efficiency
are crucial.
