Comparison of Performance of Some Classification Methods to Evaluate the Quality of Vegetables from its Morphology

Abstract

One important aspect of Data Science is its ability to classify subjects into non-overlapping groups based on one or several input variables. Several methods and algorithms are available in the literature for classifying subjects based on the values of multiple observed variables. Such classification tools are Naive Bayesian Classifiers, Logistic Regression, Discriminant Analysis, k-nearest neighbourhood etc. This paper attempts to recognise if the morphological variables, identified either through literature review or from expert opinion, can be utilised to understand the quality of vegetables. Consequently, the current researchers obtained primary data about the morphology of the vegetables through experimentation. The outcome variable is the quality of the vegetables classified as eatable or not-eatable because of worm attack. Several classification methods are then compared for the classification exercise by building the model based on the training sample and testing the performance of the models in the holdout sample. Methods of classification performance statistics like sensitivity, specificity, precision etc. are used for their comparison. The study finds that Naive Bayes and Logistic Regression models perform better for this classification exercise. For example, only eggplant (brinjal) is considered for the study.