.png){kind=icon}
.png){kind=icon}
Return
Appliance identification using Kolmogorov–Arnold networks with extended feature extraction and saliency analysis
.png){kind=icon}
PRE
AI
M
M
N
DOI：10.1007/s10489-026-07377-w.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Load identification involves detecting appliances by analyzing voltage and current waveforms. Over the years, various machine learning approaches, including ensemble methods and deep learning models such as Convolutional Neural Networks, have been proposed using diverse features and data representations. Although these models achieve high accuracy, many prioritize performance at the cost of computational efficiency, and often overlook the critical role of feature selection. In this study, we propose a Kolmogorov–Arnold Network (KAN) for appliance identification that combines model simplicity and inherent interpretability while maintaining performance comparable to state-of-the-art approaches. We introduce a structured feature extraction framework encompassing statistical, power-related, and frequency-domain features. Feature relevance is analyzed through multiple feature combinations and correlation analysis. The model is evaluated using four metrics on three public datasets: COOLL, PLAID, and WHITED, and is compared against six benchmark approaches. Using the optimal feature set, the model achieves 96% accuracy and F1 score on COOLL, 92% accuracy and 88% F1 score on PLAID, and 85% accuracy and F1 score on WHITED. The code for this study is available on our GitHub repository via the following link: https://github.com/Axiid-7/Appliance-Recognition
Keywords:
Non-intrusive load identification
Kolmogorov-Arnold Network (KAN)
Feature importance
Feature extraction
Load identification
Journal
IF:
3.5
Papers:
7.5K
Citations:
1.7W
Organization
Cited Papers