Enhancing SMS Spam Detection with Shallow Learning and Vectorization
R. Saravanan, K. Deepak · 2023
In order to protect users from unwanted and potentially hazardous communications, SMS spam classification is of the utmost importance. With a focus on the SMS Spam Collection Dataset, which is notable for its class imbalance, this research addresses the crucial challenge of classifying spam SMS. Initially, text cleaning and tokenization is performed. The effectiveness of the conventional Word2Vec and Term Frequency-Inverse Document Frequency (TF-IDF), two popular vectorization strategies, was investigated in this study. Four shallow learning methods were studied for spam/ham categorization: Logistic Regression, Support Vector Machine, Random Forest, and XGBoost. TF-IDF emerged as the best choice for vectorization, and Random Forest proved to be the best machine learning algorithm spam detection. By leveraging ensemble learning, Random Forest (RF) demonstrated robustness in facing the challenge of class imbalance, thereby boosting spam classification accuracy. RF achieved an average F1-Score (5-fold cross validation) of 89% over the Kaggle SMS Spam Collection dataset.