TY  - JOUR
AU  - Singh, Retinderdeep 
AU  - Prabha, Chander 
AU  - Noor, Ayman 
AU  - Khan, Mohammad Zubair 
AU  - Srivastava, Prakash 
AU  - Kaur, Gurjot 
PY  - 2024
TI  - Performance Analysis of Garbage Classification Using Balanced and Unbalanced Dataset with EfficientNetV2B1 Architecture
JF  - Journal of Computer Science
VL  - 20
IS  - 10
DO  - 10.3844/jcssp.2024.1310.1321
UR  - https://thescipub.com/abstract/jcssp.2024.1310.1321
AB  - This research concentrates on EfficientNetV2B1 deep learning model to classify garbage collection tasks, with both balanced and unbalanced dataset configurations. The dataset includes 7,260 images for the balanced dataset and 15,515 for the unbalanced one, both of the datasets are used to train and evaluate the model. Training and evaluation of the deep learning model with the standard performance variables: Accuracy, precision, recall, F1-score, and AUC score. The results indicate the unbalanced dataset performs excellently, with an accuracy of 95.22%, precision at 95.28%, recall of 95.22%, and F1-score of 95.21%. In contrast, the fully balanced data set yields slightly less but still excellent results: Accuracy of 91.46%, precision of 91.60%, recall of 91.46%, and F1 score of 91.43%. The test scores for loss and accuracy in both datasets are 0.4296 and 0.9522 for the unbalanced dataset and 0.60189 and 0.9146 for the balanced dataset respectively. A study with a dataset containing balanced classes is beneficial for assessing EfficientNetV2B1 deep learning performance across different classes evenly, providing a fair evaluation of the model's ability to generalize. On the other hand, a study with an unbalanced class distribution can be useful for evaluating how well EfficientNetV2B1 deep learning handles class imbalance and its performance in minority classes. Both types of studies offer valuable insights into model behavior under different data scenarios.