Plastic-Type Classification for Sorting System Based on Digital Image using Multinomial Logistic Regression with k-Fold Cross Validation

Recycling plastics has become increasingly prevalent in recent years. Recycling reduces plastic waste through the reuse of materials as opposed to their disposal. Additionally, this method aids in the reduction of pollution caused by the emission of greenhouse gases during the production of new plastic from basic materials. The initial phase of the plastic waste recycling procedure involves sorting plastic to various types of material. Accurately identifying the type of plastic is exceptionally beneficial for developing sifting systems in the recycling industry. This study aimed to test how well multinomial logistic regression with k-fold cross-validation can determine the difference among varying types of plastic. This method is a frequently employed statistical learning technique due to its generally satisfactory performance compared with alternative methods. Results showed that multinomial logistic regression performed well in identifying the type of plastic in all performance metrics. The performance average measures of five folds were 86.08% accuracy, 79.11% recall-mu, 79.08% recall-M, 89.56% specificity-mu and 89.59% specificity-M.