<![CDATA[The spread of hoax news on digital platforms demands an automatic detection system based on Neural Networks. However, the performance of these models is heavily influenced by hyperparameter configuration, where manual determination is time-consuming and prone to overfitting. This study aims to compare the performance of Genetic Algorithm (GA), Particle Swarm Optimization (PSO), and Hybrid GA-PSO in optimizing hyperparameters of Artificial Neural Network (ANN), Convolutional Neural Network (CNN), and Long Short-Term Memory (LSTM) architectures for detecting hoax news in Indonesian. Quantitative experiments were conducted using 4,132 news data, extracted through TF-IDF and Word2Vec, then trained and evaluated based on the level of classification accuracy and convergence quality. The test results show that the effectiveness of the optimization algorithm is highly dependent on the model architecture. In the baseline ANN model, PSO performed optimally with an accuracy of 96.3% and an F1-Score of 96.3% due to its ability to exploit parameters stably. In the CNN model, PSO experienced overfitting, making GA the best method with 95.4% accuracy due to its advantage in maintaining model generalization on new data. Meanwhile, for the LSTM sequential model, pure GA experienced model collapse with accuracy dropping to 49.9%. The Hybrid GA-PSO approach proved to be the most optimal method for LSTM, dominating with the highest accuracy of 96.5%, a precision rate of 97.2%, and an F1-score of 96.5%. In conclusion, algorithm hybridization is highly recommended for complex architectures to overcome the weaknesses of a single algorithm while producing sharp predictions that precisely distinguish facts from hoaxes.]]>
Copyrights © 2026
