Determining Banana Types and Ripeness from Image using Machine Learning Methods

Customers should have several benchmarks to buy banana from the market. One of them is observing each size to its ripeness. This study present a framework for determining bananas based on types and levels of ripeness from banana peel and images. We use three machine learning method, i.e. k-Nearest Neighbor ( k-NN), Support Vector Machine (SVM), and Decision Tree (DT). The banana is placed on the white background and photographed within 0.6 meters with 17 different position. The images are converted into grayscale mode and become 96x96 pixels. Principal Component Analysis (PCA) is conducted to reduce the dimensionality from 9,216 pixels to 236 pixels and 128 pixels. In this research, SVM is able to provide high accuracy compared to other methods, k-NN and DT, to determine banana types, that is 99.1%. To determine the level of ripeness, k-NN and SVM achieved the same highest result, that is 96.6%. However, SVM has the faster processing time compared to k-NN, that is 5.517s. Furthermore, SVM is also tested by using PCA 256 pixels, PCA 128 pixels, and non-PCA. The result was SVM with PCA 128 pixels was able to reduce the processing time from 5.517s to 5.492s.