Optimizing Hyperparameters in MobileNet v1 Image Classification Using Bayesian and Random Search

Fine-tuning hyperparameters plays a crucial role in boosting the accuracy and efficiency of image classification models. However, traditional approaches like grid search can be quite resource-intensive, especially when dealing with complex models. In this study, we explored two methods—Bayesian Optimization and Random Search—to optimize the hyperparameters of the MobileNet v1 model, specifically for detecting broken glass bangles using a dataset from Kaggle. We found that Bayesian Optimization converged faster, achieving an accuracy of 94.2% with fewer iterations compared to Random Search, which achieved 93.6%. Key hyperparameters, such as learning rate, batch size, and dropout rate, were optimized to improve the model's performance. Our findings show that Bayesian Optimization is more efficient for this task compared to Random Search, offering valuable insights for practitioners working with MobileNet v1. The findings provide insights into selecting effective optimization strategies to enhance model performance in defect detection tasks.