Hybrid Particle Swarm Optimization dan Multi-Layer Perceptron dalam Klasifikasi Biner

Abstract

The development of machine learning has encouraged the use of classification methods capable of producing accurate predictions on complex data patterns. Multi-Layer Perceptron (MLP) is one of the widely used models due to its ability to model nonlinear relationships. However, the training process of MLP using backpropagation has limitations, including dependence on initial weight initialization and the risk of being trapped in local minima. Therefore, an optimization method is needed to improve model performance. This study aims to implement a hybrid Particle Swarm Optimization and Multi-Layer Perceptron (PSO-MLP) method for binary classification and to analyze its performance compared to standart MLP. This study uses four datasets with different characteristics. The research stages include data preprocessing such as data cleaning, feature selection, encoding, and normalization, followed by data splitting into training, validation, and testing sets. PSO is applied to optimize the initial weights and biases in the first hidden layer of MLP, followed by training using the backpropagation algorithm. Model evaluation is conducted using a confusion matrix with accuracy, precision, recall, and F1-score metrics. The results show that the PSO-MLP method produces varying performance across datasets. The model achieves the best results on datasets with low complexity, reaching an accuracy of 99.23%, and performs fairly well on other datasets. However, for datasets with higher dimensionality or more complex patterns, the performance does not significantly improve compared to standart MLP. In conclusion, the hybrid PSO- MLP method does not consistently improve binary classification performance. Its effectiveness is highly dependent on the characteristics of the dataset used.