A comparative analysis of machine learning techniques for building cooling load prediction

This study evaluates the effectiveness of various machine learning techniques SVM, Naive Bayes, Linear Regression, and Decision Tree in estimating the cooling load for 768 samples. It explores the relationships among nine key variables, such as orientation and glazing area, to understand their influence on cooling load. The analysis reveals that these predictors can explain approximately 91% of the variance in cooling load, with an R-squared value of 0.91. The study also compares the accuracy of the algorithms, with SVM achieving the highest accuracy of 97.4%, making it suitable for high dimensional spaces. Decision Tree and Naive Bayes both achieve 96% accuracy, with the former being noted for its interpretability and the latter for its applicability to large datasets. Linear Regression, with 95% accuracy, is recognized for its effectiveness in linear relationships. The findings underscore the importance of selecting an ML technique based on the specific problem, data type, and the desired balance between complexity, accuracy, and interpretability.