Ensemble genetic and CNN model-based image classification by enhancing hyperparameter tuning.

Area of Science:

• Computer Science
• Artificial Intelligence
• Machine Learning

Background:

• Hyperparameter optimization is crucial for enhancing image classification model performance.
• Overfitting is a common challenge in deep learning, necessitating balanced model complexity and generalization.
• Existing models like CNN, RNN, AlexNet, ResNet, and VGG have limitations in achieving optimal performance.

Purpose of the Study:

• To propose an enhanced image classification model using ensemble learning and genetic algorithms.
• To fine-tune hyperparameters of Convolutional Neural Networks (CNNs) for improved accuracy and efficiency.
• To leverage the strengths of ensemble methods for superior image classification results.

Main Methods:

• Developed an ensemble genetic algorithm and convolutional neural network (EGACNN) model.
• Integrated a genetic algorithm (GA) with a CNN using stacking for hyperparameter optimization.
• Tuned CNN hyperparameters including the number of layers, kernel size, learning rates, dropout rates, and batch sizes.
• Utilized the Modified National Institute of Standards and Technology (MNIST) dataset for training and evaluation.

Main Results:

• The proposed EGACNN model achieved a highest accuracy of 99.91%.
• The ensemble CNN and spiking neural network (CSNN) model demonstrated an accuracy of 99.68%.
• EGACNN and CSNN models showed superior performance compared to standalone CNN, RNN, AlexNet, ResNet, and VGG models.

Conclusions:

• Ensemble approaches, particularly EGACNN, offer significant improvements in image classification accuracy.
• Hyperparameter optimization using GA effectively enhances deep learning model performance and reduces manual effort.
• The proposed EGACNN model represents a superior alternative for image classification tasks.