WASTE CLASSIFICATION USING TRANSFER LEARNING WITH MOBILENETV2 TO SUPPORT ENVIRONMENTAL MANAGEMENT

Muhammad Trihardyansyah (STIKOM Tunas Bangsa)
M Safii (STIKOM Tunas Bangsa)
Sundari Retno Andani (STIKOM Tunas Bangsa)

JITK (Jurnal Ilmu Pengetahuan dan Komputer)
Vol. 12 No. 1 (2026): JITK Issue August 2026

Efficient waste management requires an accurate automatic classification system for organic and recyclable categories. This study aims to develop a deep learning-based waste classification model using the MobileNetV2 architecture with a transfer learning approach. The main contribution of this study lies in the modification of the architecture through fine-tuning techniques and the integration of a 0.5 dropout layer specifically designed to address the problems of overfitting and class imbalance in waste image data. Test results show that the optimized MobileNetV2 model significantly improves classification performance, achieving an accuracy of 93%. This proposed model is proven to be more adaptive and substantially superior compared to standard architectures and comparable architectures such as ResNet.