Text Classification on Cybercrime Cases From News Articles Using Supervised Learning

Abstract

The number of cybercrime cases has increased in this country, especially after the pandemic. The nation has created numerous strategic plans, including the introduction of the Malaysia Cyber Security Strategy (MCSS), which sparked a baseline for countering cybercrime. One of the pillars is Enhancing Capacity and Capability Building, Awareness, and Education. To raise awareness effectively, the taxonomy of cybercrime must be easily understandable by the citizens. This project is to study the classification of news postings by applying supervised models that can ease the classification of cybercrime types. Five supervised models with a combination of two feature extractors were examined. The models were experimented with to evaluate their performance using a percentage split of 70:20 and 80:20. Each model is evaluated based on accuracy, F1-measure, and precision. In the experiment, Random Forest with the TF-IDF feature extractor produced the best result. Achieving an impressive accuracy rate of 94.01%, this model stands out for its precision. Naïve Bayes with the Word2vec feature extractor performed the least effectively, with an accuracy rate of 73.48%. This research focused on analyzing textual data by examining word frequency and interpreting topics based on the class labels of Cybercrime Type 1 and Cybercrime Type 2. Each class of cybercrime news uncovered the topic using latent direct allocation, which was interpreted using Chat-GPT. The analysis and the results of the classification model have been effectively visualized in the PowerBI dashboard, enhancing comprehension. To enhance future research, consider adjusting the scope of the data to focus on local Malay news for more targeted insights.