Impact of Hyper Parameter Optimization Over Performance Improvement of CNN Models

With the use of Convolutional neural Networks (CNN) in medical image processing, researchers focus on improving the accuracy of CNN model in numerous ways. One such concepts booms with optimizing the hyper-parameters of CNN models. Discovering the optimal value of hyper-parameters was a time consuming task and available optimizing techniques fails to provide the boundary of optimal hyper-parameter values. This research article was concerned with optimizing the hyper-parameter during transfer learning focusing the impact of optimal hyper-parameter value and optimization methods on CNN models. In this research we focused on selection of optimal hyper-parameter value and performed a comparative analysis of the optimization methods. Three Optimization algorithms such as Grid Search, Random Search and Bayesian Optimization were used and evaluated with accuracy metrics. It was observed that Bayesian optimization yields best CNN model with accuracy of 85.4% over other algorithms.