DETEKSI PENIPUAN PADA SOSIAL MEDIA TWITTER DENGAN METODE BIDIRECTIONAL LONG SHORT TERM MEMORY (BI-LSTM)

Abstract

This research aims to address the issue of online fraud detection in Indonesia through the implementation of the Bidirectional Long Short Term Memory (BI-LSTM) method on the Twitter social media platform. Adopting a descriptive research approach, the study seeks to comprehend user behavior, interaction patterns, and sentiments expressed on Twitter without manipulating the studied variables. Data collection involves utilizing APIs and Web Crawlers to gather information regarding online behavior. The evaluation results indicate that the BI-LSTM model outperforms the LSTM model in detecting fraudulent and non-fraudulent transactions. The BI-LSTM model demonstrates higher precision, recall, and accuracy, showcasing its superior ability to identify genuine fraudulent transactions and avoid prediction errors. These evaluation outcomes are reinforced by training and validation graphs, illustrating that the model has reached its peak performance in learning from the available training data. The conclusion drawn from this research underscores the importance of understanding the common characteristics of online fraud, utilizing the Indonesian language, and employing relevant keywords during dataset collection to develop an effective deep learning model for online fraud detection. Furthermore, employing appropriate validation methods, periodic performance evaluations, hyperparameter tuning, and dataset adjustments are crucial steps in optimizing the outcomes of the developed model. The Early Stopping technique can also be utilized to halt training when the model no longer demonstrates significant performance improvements, thereby conserving computational resources and ensuring focus on the most optimal model.

Kata kunci : Fraud Detection; BILSTM Model; Cyber Security; Machine Learning; Social Media Fraud