Improving in-home appliance identification using fuzzy-neighbors-preserving analysis based QR-decomposition

TL;DR

This paper introduces a novel appliance identification method using fuzzy-neighbors-preserving QR-decomposition for feature extraction combined with a bagging decision tree classifier, achieving high accuracy on multiple datasets.

Contribution

The paper presents a new FNPA-QR technique for feature extraction and a BDT classifier, improving appliance identification accuracy over existing methods.

Findings

High classification accuracy on three datasets

Effective feature discrimination with FNPA-QR

Enhanced appliance identification performance

Abstract

This paper proposes a new appliance identification scheme by introducing a novel approach for extracting highly discriminative characteristic sets that can considerably distinguish between various appliance footprints. In this context, a precise and powerful characteristic projection technique depending on fuzzy-neighbors-preserving analysis based QR-decomposition (FNPA-QR) is applied on the extracted energy consumption time-domain features. The FNPA-QR aims to diminish the distance among the between class features and increase the gap among features of dissimilar categories. Following, a novel bagging decision tree (BDT) classifier is also designed to further improve the classification accuracy. The proposed technique is then validated on three appliance energy consumption datasets, which are collected at both low and high frequency. The practical results obtained point out the outstanding classification rate of the time-domain based FNPA-QR and BDT.