Effect of TF-IDF Extraction and Application of SMOTE on Model Performance in Detecting Spam Email
Muhammad Fadli, Vannes Wijaya, Muhammad Rizky Pribadi, Wijang Widhiarso · 2023
Email is one of the communication media that sends digital letters electronically using the internet. Email abuse frequently occurs, and one example of email abuse is email spam. There are many approaches that can be used to help filter email spam, but crucial techniques in text classification, such as email spam classification, include text feature extraction, as well as dataset balancing techniques like SMOTE, which can improve model performance. This research has the purpose to determine the extent of the influence of TF-IDF text feature extraction and SMOTE implementation on the performance results of the classification model. The models to be used are Naïve Bayes, Deep Learning, and K-Nearest Neighbors. Based on the results of the conducted testing, the utilization of TF-IDF and SMOTE techniques had a significant impact on the classification model, particularly on the Deep Learning model, where the accuracy increased by 44.14% and the f1 score improved by 73.75%.