Systematic Comparison of Machine Learning Model Accuracy Value Between MobileNetV2 and XCeption Architecture in Waste Classification System - Research report

Aug 4, 2026

Upload by Miles

Daily garbage creates environmental pressure because some waste types decompose slowly, increasing pollution risks. Inorganic waste with recycle value—such as cardboard, metal, paper, glass, plastic, rubber, and packaging—benefits from automated sorting. This study builds a waste classification system using Convolutional Neural Networks, focusing on the influence of CNN architecture, optimizer, and learning rate on accuracy. MobileNetV2 with SGD (learning rate 0.1) reaches 86.07% accuracy, while Xception with Adam (learning rate 0.001) reaches 87.81% accuracy.

LanguageEnglish

CategoryResearch & Report

FormatPDF

Pages8

Reading Time20 min

AccessFree to download

1. Introduction

2. Materials and Methods

2.1. Machine Learning

2.2. Convolutional Neural Network

2.3. Convolutional Layer

2.4. MobileNetV2

2.5. Xception

Original Article Systematic Comparison of Machine Learning Model Accuracy Value Between MobileNetV2 and XCeption Architecture in Waste Classification System Yessi Mulyani1, Rian Kurniawan1,*, Puput Budi Wintoro1, Muhammad Komarudin1, and Waleed Mugahed Al-Rahmi2 1 Department of Electrical Eengineering, Faculty of Engineering, Universitas Lampung, Jl. Prof. Sumantri Brojonegoro no.1, Bandar Lampung 35145, Indonesia 2 Faculty of Social Sciences and Humanities, School of Education, Universiti Teknologi Malaysia, UTM Johor Bahru Skudai 80990, Malaysia * Correspondence: [rian.kurniawan1018@gmail.com](rian.kurniawan1018@gmail.com) [Received: 12 August 2022](Received: 12 August 2022); [16 October 2022](16 October 2022); 30 December 2022 Abstract. Garbage generated every day can be a problem because some types of waste are difficult to decompose so they can pollute the environment. Waste that can potentially be recycled and has a selling value is inorganic waste, especially cardboard, metal, paper, glass, plastic, rubber and other waste such as product packaging. Various types of waste can be classified using machine learning models. The machine learning model used for classification of waste systems is a model with the Convolutional Neural Network (CNN) method. The selection of the CNN architecture takes into account the required accuracy and computational costs. This study aims to determine the best architecture, optimizer, and learning rate in the waste classification system. The model designed using the MobileNetV2 architecture with the SGD optimizer and a learning rate of 0.1 has an accuracy of 86.07% and the model designed using the Xception architecture with the Adam optimizer and a learning rate of 0.001 has an accuracy of 87.81%. Keywords: classification, convolutional neural network, machine learning, MobileNetV2, Xception 1. Introduction Garbage generated every day can be a problem because some types of waste are difficult to decompose so they can pollute the environment [1,2]. Waste consists of two types, namely organic waste and inorganic waste. Organic waste is waste that comes from the remains of living organisms, while inorganic waste comes from non-living organisms. Waste that can potentially be recycled and has a selling value is inorganic waste, especially cardboard, metal, paper, glass, plastic, rubber and other waste such as product packaging [1] . Various types of waste can be classified using machine learning models. The machine learning model used for classification of waste systems is a model with the Convolutional Neural Network method. The model with this method will recognize garbage images by extracting image features and recognizing patterns according to the labels on the training data [2,3] . In the last few decades, deep learning has become a powerful tool. This is evidenced by its ability to handle large amounts of data and be able to recognize patterns from the data it manages. One of the popular algorithms for handling large amounts of data is the Convolutional Neural Network [3]. The Convolutional Neural Network architecture used to design machine learning models plays an important role. The more precise the choice of Convolutional Neural Network architecture, the better the accuracy of the model made to predict the garbage image. In addition to good accuracy, each architecture has a different size, parameters, and cost (CPU/GPU) . The Xception architecture isan architecture that has high accuracy, small size, and fewer parameters than some other architectures, so that models trained using this architecture will be used effectively and efficiently to predict images. The MobileNetV2 architecture is an architecture that has high accuracy with the smallest size and fewest parameters compared to other architectures, so that models trained using this architecture will be efficient in terms of training time and can predict the image quite well [4]. 2. Materials and Methods 2.

FAQs

What problem does the document address in waste management?

It addresses the environmental impact of daily garbage, especially waste types that are difficult to decompose, and the need to classify recyclable inorganic waste for better handling.

Which machine learning approach is used for the waste classification system?

A Convolutional Neural Network (CNN) is used to recognize waste images by extracting image features and learning patterns from labeled training data.

How do the two CNN architectures compare in accuracy?

MobileNetV2 with SGD (learning rate 0.1) achieves 86.07% accuracy, while Xception with Adam (learning rate 0.001) achieves 87.81% accuracy.

Aug 4, 2026

Upload by Miles