Spam Detection on YouTube Comment Section: Comparison Between Deep Learning and Machine Learning Methods

The comment sections of YouTube are increasingly targeted by spam generated by automated bots, leading to harmful and misleading content. This study evaluates the performance of classical machine learning (ML) and deep learning (DL) models in detecting spam comments using a balanced dataset of 2,519 entries. Data preprocessing included tokenization, lemmatization, and stopword removal, followed by word embedding techniques such as TF-IDF and Word2Vec. Machine learning models, including Logistic Regression, Random Forest, Support Vector Machines, and FastText, were compared against deep learning models like Long Short-Term Memory (LSTM) and Gated Recurrent Unit (GRU). Among these, FastText achieved the highest accuracy (96.00%) and ROC AUC (0.9597), outperforming all other models. The GRU model emerged as the best deep learning model, with an accuracy of 91.60% and a ROC AUC of 0.9151. The findings highlight that ML models, particularly FastText, demonstrate superior accuracy and efficiency for this task compared to DL models. Future work should investigate larger datasets, advanced architectures like Transformers, and hybrid approaches to further enhance spam detection in YouTube comment sections.