Klasifikasi Pesan Spam Menggunakan Algoritma Recurrent Neural Network (RNN)

Abstract

The increasing use of communication medias in Indonesia is fertile ground for spammers to send their messages. These messages are usually done for promotional activities and dissemination of public information for the community, but sending these mass messages must be in accordance with the consent by the user because otherwise these messages can be said to be spam messages. Spam messages can also pose security risks such as theft of personal data and also the spread of malware sent through links. Not a few Indonesians are deceived by spam messages due to a lack of digital literacy and understanding of fraud modes through these mobile messages. This research was conducted with the aim of classifying spam messages based on their types, namely non-spam, fraud, and promotional messages using the Recurrent Neural Network algorithm. The research was conducted using 1200 messages that went through the Preprocessing stage consisting of Case Folding, Normalization, Stopword Removal, and Stemming. Then passed the Tokenization stage and TF-IDF and N-gram feature extraction and Chi Square feature selection. The accuracy of the developed model is 0.92 and the loss is 0.53. After several evaluations, it can be concluded that the algorithm used by the author can classify spam messages based on their type, namely non-spam, fraud, and promotion messages with good performance.