This paper explores the use of Convolutional Neural Networks (CNNs) for solid waste
recognition, using Google Teachable Machine. A CNN model was trained on 13,745 images
from Kaggle datasets to classify waste into four categories: Dangerous, Recyclable,
Organic, and Non-Recyclable. Evaluation on 80 test images showed an overall accuracy
of 82.5%, with the highest performance in the Dangerous, Organic and Non-Recyclable
categories, while Recyclable waste had the highest misclassification rate. A confusion
matrix analysis revealed that Recyclable waste was often misidentified as other categories.
A comparison between manual testing and Google Teachable Machine’s accuracy reports
showed consistent classification trends, reinforcing the importance of real-world
validation. Accuracy and loss per epoch graphs confirmed stable training. The findings
highlight the potential of AI in waste management and suggest improvements such as
dataset expansion and real-time image augmentation.
