Comparative Study of Classification of Split Bulk Grams Using Different Significant Feature Selection Techniques
K.R. Pushpalatha, Asha Gowda Karegowda · International Journal of Applied Research on Information Technology and Computing · 2018
This paper presents the comparative study of similar looking split bulk grams using two major feature extraction methods, Gray Level Co-occurrence Matrix (GLCM) and Gray Level Run Length Matrix (GLRLM). For comparison 22 GLCM features, 11 GLRLM features and 33 combined features (GLCM+GLRLM) are extracted from five different types of similar looking split bulk grams. As a part of significant feature selection step, two approaches, namely Correlation-based Feature Selection (CFS) evaluator and consistency-based subset (CSE) evaluator have been applied separately with six different searching methods. The evaluation of significant features selected is done using six different classifiers. Result proves that feature set selected by the GLCM is better when compared to that with GLRLM features. Further, CFS with Best First Search (BFS) identified features (from combined GLCM+GLRLM)resulted in best accuracy of 88% for Multilayer Perceptron classifier. In addition, CSE with Genetic Algorithm (GA)-based search identified features (from combined GLCM+GLRLM) resulted in best accuracy of 87% for Multilayer Perceptron classifier.