Multiclass Classification of Dry Bean Grains Using Machine Learning Techniques

The bean (Phaseolus vulgaris) is a crucial crop for global food security; thus, grain quality plays a meaningful role in the supply chain. Traditional classification involves the manual evaluation of parameters such as length, width, weight, and inaccuracies. However, this manual approach presents challenges in terms of efficiency and subjectivity, as it can be slow due to human errors. This study aims to identify bean grain varieties in terms of their shape by using a machine learning algorithm approach. A database of 3,820 grains from three types—INIAP-420 (1,320), INIAP-422 (1,200), and INIAP-425 (1,300)—was used, containing 7 morphological shape variables. For the analysis of these variables, classification models were created using Random Forest (RF), Support Vector Machine (SVM), Multilayer Perceptron (MLP), and k-Nearest Neighbors (k-NN) with 10-fold cross-validation, and performance metrics were compared. Overall correct classification rates were determined to be 86.65%, 93.12%, 90.39%, and 87.58% for RF, SVM, MLP, and k-NN, respectively. The SVM classification model, which showed the highest accuracy results, classified the bean types INIAP-420, INIAP-422, and INIAP-425 with 97.72%, 98.33%, and 97.46%, respectively. Based on the achieved performance measurement values, it can be concluded that the study successfully classified bean grain varieties.