|
|
 |
| |
|
|
|
Development of an AI-Based Communication Fraud Detection System |
|
|
|
PP: 953-963 |
|
|
|
doi:10.18576/amis/190419
|
|
|
|
Author(s) |
|
|
|
D. N. Kuanyshbay,
A. G. Serek,
A. A. Shoiynbek,
K. R. Sharipov,
T. A. Shoiynbek,
B. A. Meraliyev,
M. A. Meraliyev,
|
|
|
|
Abstract |
|
|
| Traditional rule-based spam filters have proven insufficient against the increasing fraudulent SMS and messaging platform activities thus driving the need for AI-based detection systems. This research compares five traditional machine learning models including Na ̈ıve Bayes, Logistic Regression, Support Vector Machines (SVM), k-Nearest Neighbors (KNN) and Decision Trees for SMS spam detection using TF-IDF feature extraction methods. The SMS Spam Collection Dataset contained 13.4% spam messages which served as the basis for training and testing the models. The combination of SVM with TF-IDF produced the best results by achieving an F1-score of 0.96 and perfect precision of 1.00 together with a recall of 0.92 for identifying spam messages. The F1-score reached 0.90 for Logistic Regression but Na ̈ıve Bayes reached 1.00 precision at the cost of 0.75 recall. The KNN model demonstrated weak performance because its spam F1-score reached only 0.56 while the Decision Tree model produced an F1-score of 0.87. The ROC- AUC scores demonstrated that SVM (0.99) and Logistic Regression (0.99) outperformed all other classifiers. The obtained results show that simple yet interpretable models can deliver high accuracy in spam detection and establish a solid base for implementing AI-based fraud detection systems.
|
|
|
|
|
|
|
|
