Multivariate Classification of Dry Beans using Pipelined Dimensionality Reduction Technique

Machine learning methods have been employed frequently in the agricultural sector for crops analysis. These artificial intelligence-based models facilitate the understanding about crop sub species, their diseases and other significant issues concerning their yield. It helps in classification of several species of similar crops and also to identify specific sub-species among several others. The machine learning models can also be utilized to classify different types of beans, grains and pulses. It greatly assists in the evaluation of high-quality grains and their uniformity in size. Many features define each variety of the beans but not all the features equally contribute in the process of classification. In this paper, a Pipelined Linear Discriminant Analysis methodology has been used to remove insignificant features and estimators were applied for classification. Seven machine learning models were investigated and the overall prediction accuracy was 92%. Support Vector Machine, Random Forest and Multilayer Perceptron had higher accuracy of 93%. This methodology also efficiently overcomes the problem of overfitting and at the same time increases the overall performance of the models. We further investigate the impact of dimensionality reduction on the dataset.