|
|
 |
| |
|
|
|
A Heterogeneous Ensemble Learning Framework-based Binary Genetic Algorithm for Predictive Maintenance of HVAC Systems in Medical Facilities |
|
|
|
PP: 181-213 |
|
|
doi:10.18576/amis/200113
|
|
|
|
Author(s) |
|
|
|
Bilal Bataineh,
|
|
|
|
Abstract |
|
|
| The operational integrity of Heating, Ventilation, and Air Conditioning (HVAC) systems is critical in medical facilities, directly impacting patient safety, infection control, and significant operational costs. Traditional maintenance strategies are often reactive and inefficient, creating a need for more intelligent, proactive solutions. To address this issue, this research proposes a robust, heterogeneous ensemble learning framework. A Binary Genetic Algorithm (GA) is employed to automatically select the optimal subset of weak learners from a pool including Random Forest, Gradient Boosting, SVM, and AdaBoost, among others, to maximize predictive performance. The final optimized ensemble utilizes a soft-voting strategy for prediction. The framework’s performance is rigorously validated using a repeated 10-fold cross-validation methodology on real-world HVAC sensor datasets collected from ten medical facilities in Jordan, ensuring the stability and generalizability of the results. Key findings indicate that the proposed GA-optimized ensemble achieves an average macro-averaged F1-score of 77.35% (±0.010) across datasets, outperforming the naive ensemble’s 64.65% (±0.011), with overall accuracy reaching 97.50% (±0.009) vs. 91.82% (±0.018) due to class imbalance favoring majority classes (Excellent/Good, comprising 70% of instances). It is demonstrated that the optimized subset of models often outperforms a naive ensemble of all learners, showcasing improved efficiency and model synergy..
Keywords: |
|
|
|
|
|
|
|
