Hyperspectral Band Selection based on Decision Tree Algorithm in Beeswax Identification on Rome Beauty Apple

Abstract

Machine learning has been a big help to perform specific tasks by learning the data and improving the results. The way this system works was based on pattern recognition and computational algorithms. Some classification algorithms go through a process called feature selection or dimensionality reduction. This process was used to minimize the number of features used. In this study, the process was observed through a hyperspectral image to identify beeswax on Rome Beauty apples and to define the essential variables on the wavelengths. The hyperspectral image was acquired on a wavelength ranging from 400 to 1000 nm. The spatial and spectral data of the image can be obtained through this technique. Thus the reflectance profile from the object was used to classify the nonwaxed apple and the waxed apple based on the variable importance. Compared to the accuracy of the support vector machine model, the accuracy of the decision tree model shows a better outcome with 81.25% correct predictions from 48 testing data. In the decision tree model, there are 13 essential variables on 13 features (wavelength) that was used by the classifier to get the best result.