Performance of Information Gain and PCA Feature Selection for Determining Ripen Susu Banana Fruits

Abstract

Susu banana fruits has a uniqueness, where is the difference of slightly ripe and ripe susu banana at the ripen stage is perfectly difficult to distinguish visually because both have almost the same yellow color. Therefore, this study performed identification using a fruit image-based computer vision to replace the human visual. The almost similar characteristics of susu banana at slightly ripe, ripe and riper stage were selected to get a dominant character that has a high influence. The ability of information gain (IG) and principal component analysis (PCA) and combined IG-PCA features selection was evaluated to determine the influence of correlation and probability of each feature on each class. Tests were conducted on clean-peeled and spotted peel susu banana with 3 levels of ripeness at the ripen stage to determine the impact of IG, PCA and combined IG-PCA on classification using extreme learning machines. The test results showed that the use of PCA in the clean-peeled with natural curing (group1) and spotted peel with chemicals curing (group3) was better than IG. In the group1, PCA also outperformed combined IG-PCA, but in the group3 the combined use of IG-PCA was better than IG and PCA. Although the use of feature selection at spotted peel with natural curing (group2) was resulted the lower accuracy, overall, the tests showed that the selected of dominant features in the classification could increase the recognition accuracy. The proposed method also proved could be used as an alternative in determining the ripen of susu bananas.