<![CDATA[Email is a means of communication within internal networks and the internet for the exchange of information. Email is still used today because of its ease of use. However, with the increase in the number of incoming emails, the problem of spam has arisen, requiring effective methods for detecting spam so that users can manage their email more efficiently and avoid potential fraud and disruption. This study aims to analyze the thematic and linguistic patterns of email messages based on their content using text classification techniques with the Multinomial Naive Bayes algorithm, which is believed to have good accuracy in detecting spam emails. The research consists of collecting a dataset related to Indonesian-language spam emails, preprocessing the data, training the model by dividing it into two scenarios (with and without stemming), and evaluating the model. Features from the email text will be converted into numerical representations using the Bags-of-Words method. Classification performance evaluation is carried out using accuracy, precision, recall, F1-Score, and confusion matrix metrics. Experimental results demonstrate that the Multinomial Naive Bayes model without stemming achieved the highest performance with an Accuracy of 92.5%, Precision of 91.0%, and F1-Score of 91.7%. These findings indicate that stemming in short texts like spam emails eliminates crucial semantic features (affixes) characteristic of spam. This study contributes to providing optimal pre-processing recommendations for Indonesian short text classification.]]>
