Abstract

This research intends to use machine learning techniques to estimate the heating load class of residential buildings, allowing for the design of more energy-efficient structures. The dataset utilized in the analysis comprises of seven hundred residential building examples, each with variables such as relative compactness, surface area, roof area, overall height, orientation, glazing area, and glazing area distribution. The target variable, heating load class, is classified as low, medium, or high. The analysis begins with a thorough overview of the major variables, which includes summary statistics, correlation analysis, and visuals. Decision tree models are built in two ways: using the complete dataset and adopting a 10-fold cross-validation strategy. Additionally, a Neural Net Multi-Layer Perceptron (NN-MLP) model is run and compared to the decision tree models. The results show that all models have high accuracy rates, with the NN-MLP model having the best at 99.57%. Comparative research indicates that the NN-MLP model outperforms decision tree approaches in forecasting heating load classes. However, both models perform well in accurately categorizing heating load classes. Overall, this work emphasizes the effectiveness of machine learning algorithms in forecasting heating load classes, as well as their potential for improving the energy efficiency of residential structures.