Refining Parameter Values in Convolutional Neural Networks Using Weight Modification Particle Swarm Optimization for Enhanced Coral Reef Condition Classification

Indonesia has a vast marine area with numerous coral reefs located at the center of the global coral triangle, which is crucial for aquatic life.However, these coral reefs face severe threats from climate change, pollution, rising sea temperatures, and human activities.Currently, 85% of Indonesia's coral reefs are at risk.Therefore, monitoring and preserving coral reefs are essential for the sustainability and health of the ocean.Technology such as Convolutional Neural Networks (CNN), which monitor and classify coral reef conditions, is highly beneficial in addressing these challenges.This study shows that the optimization method based on swarm optimization, namely Particle Swarm Optimization (PSO) influenced by declining weighting, effectively improves the performance of CNN by determining the best values for each parameter in the CNN for classifying coral reef conditions.The optimized parameters include Epoch, Dropout, Hidden Size, Learning Rate, Activation Function, and Optimizer.The experimental results indicate that the best parameter values produced by the proposed are Epoch of 9, Dropout of 0.2, Hidden Size of 256, Learning Rate of 0.002, Activation Function as ReLU, and Optimizer as Adam.The proposed shows more consistent and stable performance compared to the Original PSO and Original GWO (Grey Wolf Optimizer) methods, achieving a maximum training accuracy of 82.23% and a testing accuracy of 79.74%.This indicates that the proposed can be considered more reliable and effective for parameter optimization in classifying coral reefs using CNN.