Evaluation of Plastic Waste Classification Systems

Abstract

Aim: A comparison study of three classifiers was carried out to identify the best classifier which can be utilized to automate and enhance the manual process of identifying and sorting plastic waste. Place and Duration of Study: Department of Computer Science and Engineering, Obafemi Awolowo University, between April 2014 and September 2015. Methodology: Collection of plastic wastes data from purposely selected disposal sites was done and the distinguishing characteristics (average spectrum power and shape area) of those plastic wastes were computed and used as feature data. The three classifiers designed using machine leaning and statistical techniques were implemented in the MATLAB environment. The classifiers are Fuzzy inference system, multi-layer perceptron and linear discriminant analysis. The efficiency of the three classifiers was compared using mean square error, mean absolute error and receiver operating characteristics. Results: It was observed that the classifier designed using artificial neural network had the lowest mean absolute (0.07) and mean square error (0.07), compared to other classifiers. More so, the neural network model had the highest correct classification accuracy of 92.98% as against 87.72% and 75.44% recorded for fuzzy inference system and linear discriminant analysis, respectively. Conclusion: The study has successfully classified plastic waste data using the spectrum power from the sound signal produced from plastics and the plastic's shape area. Thus, confirming that sound wave signal from plastic could be utilized as feature data in plastic waste identification.