Klasifikasi Jenis Sampah Berbasis Convolutional Neural Network dengan Optimasi Hyperparameter Tuning Arsitektur Mobilenet

Abstract

Waste management in Indonesia faces significant challenges with an increasing volume reaching approximately 175,000 tons per day. Public awareness of the dangers associated with improper waste disposal remains low, as many continue to litter indiscriminately. Waste sorting is the most effective method, involving separation based on waste types. Manual waste sorting is nonetheless inefficient, as it requires large spaces, substantial labor, and is prone to errors. This study aims to develop a waste classification model based on Convolutional Neural Network (CNN) with hyperparameter tuning optimization for the MobileNet architecture. The research adopts the Cross-Industry Standard Process for Data Mining (CRISP-DM) methodology and utilizes datasets from three waste categories organic, inorganic, and hazardous and toxic materials (B3) sourced from open Kaggle datasets. Model training was conducted using the MobileNet architecture with hyperparameter tuning optimization and resulting in optimal parameters Adam optimizer, learning rate of 0.01, batch size of 32, and 256 neurons. The results show that the model achieved 96% accuracy before optimization which increased by 2% to 98% after optimization. The model demonstrated high computational efficiency with the number of floating-point operations per second reaching 1.146 GFLOPS.