Optimal Feature Selection for Classification of Electricity Consumption

Feature selection is the essential process to obtain the best feature vectors in pattern recognition system. These feature vectors contain information describing the original data’s important characteristics. In this research, a framework based on factor analysis technique namely the Principal Component Analysis (PCA) is performed to determine the best features extracted from the daily load curve prior to clustering process. The rules of thumb applied include Bartlett’s test of sphericity, Kaiser-Meyer-Olkin (KMO) measure, Kaiser Criterion, Scree test along with Varimax approach. Accordingly, KMO as well as Bartlett’s test suggested the data factorability is significant. Furthermore, Kaiser Criterion and Scree test together with component matrix approach implied that the first two most significant factor must be retained whilst Varimax approach confirmed that clustering analysis should comprise of the entire load curve values. Upon selection of features, the capability of fuzzy clustering in classifying these features attained from 247 feeders in a particular distribution network is examined. Initial results demonstrated the effectiveness of feature selection process and the potential of fuzzy clustering in particular the fuzzy cmeans (FCM) in classifying electrical energy consumption.