<![CDATA[The high fluctuation in electricity usage and the complexity of historical data distribution in Internet of Things (IoT)-based smart home environments frequently trigger significant uncertainty in household energy efficiency management. Without an intelligent predictive system, energy consumption surges become exceedingly difficult to control accurately. Therefore, this study aims to develop a high-accuracy predictive classification model architecture to map energy consumption levels into three main categories (Low, Medium, High) by leveraging the Random Forest algorithm. The experimental process was conducted systematically, beginning with the evaluation of a baseline model, followed by an optimization phase integrating the GridSearchCV method based on 5-fold cross-validation for exhaustive hyperparameter tuning on the IoT Smarthome Energy Dataset. Performance was comprehensively evaluated using a confusion matrix and gap accuracy analysis to ensure model robustness. Experimental results proved that the hyperparameter optimization process successfully boosted the global accuracy rate significantly from an initial baseline of 98.92% to 99.46%, while effectively reducing the number of misclassifications. Furthermore, the feature importance analysis provided a transparent scientific interpretation, where the future_consumption_kWh feature was identified as having the most dominant influence at 61.75% on the model's decision structure. In conclusion, this optimized Random Forest architecture is proven to be highly robust and ready to be implemented as a real-scale automatic energy-saving instrument.]]>
Copyrights © 2026
