A novel feature selection technique for number classification problem using PNN—A plausible scheme for boiler flue gas analysis

Abstract

A number of industrial problems involve accruing data from sensor like devices for further analysis more specifically in applications such as boiler flue gas analysis, computer vision, etc. Several feature selection techniques have been utilized by researchers for data conditioning, such as sequential search technique, branch and bound technique, best individual selection technique, etc. This study reports on the plausible solution for ascertaining the composition of gases during complex boiler flue gas data analysis by taking a number classification problem as a model. For this purpose an indigenously developed arithmetic residue (AR) scheme has been devised as a feature selection technique. For the purpose of classification of data (number of classes of gases), a probabilistic neural network (PNN) has been implemented and its classification capability has been analyzed first for the data acquired from ORSAT analyzer and then for the data from KANE ® analyzer.