Enhancing Machine Learning Performance in Cyberbullying Detection through Hyperparameter Optimization

Cyberbullying cases are experiencing a significant increase, particularly in the post-Covid-19 pandemic era. Machine learning plays a crucial role in detecting whether social media content constitutes cyberbullying by learning from data. Previous research has resulted in an accuracy of 92.58%. Therefore, this research aims to improve the performance of machine learning in cyberbullying detection through hyperparameter optimization using Grid Search and Random Search. The second objective is to compare the training time between Grid Search and Random Search. Machine learning algorithms that were used included Support Vector Machine (SVM), Logistic Regression, Multinomial Naïve Bayes, Decision Tree, Random Forest (RF), and AdaBoost. These algorithms were subjected to hyperparameter optimization using grid search and random search. The analysis results showed that text preprocessing significantly affected the improvement of machine learning performance, with the highest F-1 Score Macro Average of 0.94 achieved in RF. Furthermore, hyperparameter optimization elevated the performance of SVM from 0.93 to 0.94. Meanwhile, the fastest time to get the best parameters is obtained with Random Search with an average time of 286.571 seconds and Grid Search with a slower time of 5231.2055 seconds. These results were accessible at uns.id/winarHO.