Comparative Analysis of Linear Regression and Decision Tree for Household Energy Consumption Prediction

Abstract

Household energy consumption continues to increase alongside the rapid adoption of electrical appliances and evolving lifestyles, contributing to higher global energy demand and environmental pressure. Despite the growing use of machine learning for energy prediction, many studies implicitly assume the superiority of complex non-linear models without adequately considering the characteristics of the underlying data. This study addresses this gap by conducting a comparative analysis between Linear Regression and Decision Tree Regressor for predicting household electricity consumption using a small-to-medium scale dataset. The dataset consists of 114 records with features including appliance type, usage duration, and month of use, divided into 80% training and 20% testing data. Model performance is evaluated using Root Mean Square Error (RMSE) and coefficient of determination (R²). The results show that Linear Regression outperforms Decision Tree, achieving an RMSE of 1.1552 kWh and R² of 0.96, while Decision Tree yields an RMSE of 7.3485 kWh and a negative R², indicating poor generalization on this dataset. Feature importance analysis reveals that usage duration (waktu_pemakaian) is the most dominant predictor, contributing 88.31% to the model's decisions, followed by month of use (11.69%), while appliance type contributes negligibly (0.00%). These findings demonstrate that for datasets with relatively simple linear structures, linear models can provide more stable and accurate predictions than non-linear approaches. This study contributes empirical evidence on the importance of aligning model selection with data characteristics in household energy prediction.