Tags:anti phishing solution, anti smishing, Artificial Intelligence, artificial intelligence classifier, artificial intelligence model, bagging classifier, digital signature, hash function, hash value, Lightweight cryptographic, link based feature, Link-based, machine learning, Phishing Attack, random forest classifier, recall f measure, Smishing, smishing message and SMS
Abstract:
The utilization of Short Message Service (SMS) in smartphones has expanded quickly during recent years. Roughly billions of instant SMSs are sent each hour worldwide, so the expanding utilization of these messages makes it profoundly suitable for assailants. Smishing is one of the cybercrime types that combine SMS and phishing in which trespassers send SMS containing suspicious links and content to the victims. Usually, smishing directs users to harmful websites to silently download malware or show interfaces like legitimate sites to attract users to fill their sensitive information, i.e., passwords. This study proposes two automated models for detecting smishing by employing Lightweight cryptographic and arti-ficial intelligence (AI) techniques. For the first model, and since there is no offi-cial validation stage in SMS construction, we developed our first contribution us-ing a lightweight cryptographic SMS model to protect SMS and detect possible phishing or fake content. We select the standard, popular signature schemes, and evaluating the performance according to time and signature size overhead. The re-sults showed the proposed model's effectiveness to guarantee the main security goals, authentication, integrity, and non-repudiation. Our second model deals with legacy SMS, which doesn't follow the validation stage, so we collected the SMS dataset for an AI model and manually labeled their classes as phishing or legitimate according to their link content. We preprocessed the SMS dataset to ex-tract the link-based distinguishing features, and we applied feature selection tech-niques to obtain the best performance. We experimented with two artificial intelli-gence classifiers and evaluated their detection accuracy results using accuracy, precision, recall, and F-score metrics. Random forest classifier yielded the best detection results with an accuracy of 97%.
Lightweight Cryptographic and Artificial Intelligence Models for anti-Smishing