A Decision Tree Classification Model for Cocoa Beans Quality Evaluation based on Digital Imaging
Cocoa is an important export commodity and a major foreign exchange earner for producing Countries. Cocoa beans of good qualities have more economic value than defective ones. The chocolate obtained from the good cocoa gives an aroma that is sought after. Defects in cocoa beans are detected and revealed through a cut test procedure usually carried out by the Federal Produce Inspectors. In this research, the use of a supervised machine learning approach, a decision tree, for the classification of cocoa beans into good, slatty, mouldy and weevilly beans was proposed. A digital camera was used to capture images of 300 cut cocoa beans and the images were saved in the RGB colour format on a computer system. Colour and statistical features were extracted from these images by moment analysis of the colour channels (Red, Blue and Green). Fourteen features were totally extracted and saved in a feature database. Classification and regression tree (CART), a decision tree algorithm, was employed for the classification and three splitting criteria were tested: gini diversity index (gdi), twoing and deviance. The model was evaluated by its average prediction accuracy and the receiver operating characteristics (ROC) curve for each class was plotted. The maximum classification accuracy of 89.2% was recorded from the gdi splitting criteria. With this high classification accuracy, it was concluded that the decision tree machine learning algorithm is effective in the classification of dried farm produce like cocoa.
With this high classification accuracy, it can be concluded that decision tree machine learning algorithm is effective in the classification of dried farm produce like cocoa.