Performance evaluation of SVM-Based appliance classification for NILM using the COOLL dataset

Non-intrusive load monitoring (NILM) is a method used to estimate the energy consumption of individual appliances from the aggregated electricity usage of a building without requiring dedicated sensors for each appliance. Compared with intrusive load monitoring, NILM offers a more cost-effective and scalable solution for energy management. This paper investigates appliance classification performance using the Controlled On/Off Loads Library (COOLL) dataset, which contains measurements from 42 appliances grouped into 12 appliance classes. Three feature sets were evaluated, namely steady-state power, transient-state power, and a combination of steady-state and transient-state power. Support Vector Machine (SVM) classifiers with linear, quadratic, and cubic kernels were applied to assess the effectiveness of different feature representations. The results show that the Cubic SVM with the combined steady-state and transient-state feature set achieved the highest test accuracy of 88.1%. In comparison, the best performance using steady-state features alone was 66.7% with the Quadratic SVM, whereas transient-state features alone produced lower accuracies ranging from 36.3% to 37.5%. These findings indicate that combining steady-state and transient information improves the discriminative capability of the features and enhances appliance classification performance. Therefore, the selection of suitable feature sets and SVM kernels plays an important role in improving NILM classification on the COOLL dataset.