Classification of Roasted Coffee Beans with Principal Component Analysis and Random Forest

Abstract

Roasted Robusta coffee beans originating from Toraja and Ciwidey, Indonesia have the same shape, color and characteristics of the coffee beans. There is often an error in recognizing coffee bean varieties just by looking with the naked eye. In terms of image classification, Random Forest has stable performance and performance. This research aims to classify coffee beans based on images of roasted coffee beans using the Random Forest and PCA algorithms. The steps taken were collecting Robusta and Toraja datasets which were taken manually and given 2000 labels. Data collection was carried out using a light intensity of 300 lux and a distance of 15 cm. The data was preprocessed, variables were reduced using PCA and Random Forest in forming the model. Testing uses 1600 training data and 400 test data to measure RF performance by changing the parameters of the number of data sets and the n value in Random Forest with the highest accuracy value of 98.05% with a value of n = 20 and 360 data tested.