A Generalizable Random Forest Model for Hidden Failure Prediction in Hemodialysis Equipment

چکیده

Hemodialysis machines are critical life-support devices whose undetected failures can compromise patient safety and increase healthcare costs. Most existing predictive maintenance models are developed and tested on single datasets from machines of the same generation, leaving a critical gap in understanding their performance on older equipment with different failure patterns.This study aims to develop and validate a generalizable Random Forest-based predictive model for detecting hidden failures in hemodialysis machines. A key novelty is the cross-generational validation design, which tests whether a model trained on modern machines can generalize to significantly older equipment without retraining.The model was trained on Dataset A: 14 modern machines (installed 2021, 25 inspection cycles, 47 failures) using three key sensors (blood pressure, conductivity, blood leak detector). External validation was performed on Dataset B: 6 legacy machines (installed 2006, 55 inspection cycles, 84 failures, 18 years of operational data) without any retraining—a rigorous test of generalizability across 15 years of technological difference. Performance was evaluated using accuracy, precision, recall, F1-score, and AUC.Internal validation (Dataset A) achieved 92.3% accuracy (AUC = 0.97). External validation on the independent legacy dataset (Dataset B) maintained 89.1% accuracy (precision: 88.9%, recall: 84.2%, F1: 86.5%). The model also predicted critical degradation in 4 retired machines with an average lead time of 9.2 months and 87.5% alert-level accuracy. The proposed model demonstrates strong generalizability across different equipment generations, confirming that a model trained on modern devices can effectively predict failures on 18-year-old machines. This cross-generational validation approach provides a benchmark for future research in medical equipment prognostics and offers a practical tool for hospital predictive maintenance programs.