Classificação de arroz: um estudo de caso utilizando a rede neural MLP

Abstract

This work aims to analyze the performance of the Multilayer Perceptron (MLP) neural network model in the classification of rice varieties. The database, obtained from the Kaggle platform, presents 18185 observations of the Jasmine and Gonen varieties. To implement the MLP classification model, in the Python programming language, the MLPClassifier module of the Scikit-Learn library was used. Scikit-Learn is a library developed specifically for practical application of machine learning. Model training involves hyperparameter selection and cross-validation to avoid overfitting. The resample function of the Scikit-Learn library was used to balance the classes. The performance of the MLP model was verified through the accuracy and AUC (Area under the ROC Curve) metrics. Experimental results demonstrated that the MLP model presented excellent results: accuracy = 99% and AUC = 0.9869.