A Phishing Detection Model for Financial Web-Based Applications
Phishing is a prevalent cybercrime in which attackers deceive internet users into divulging sensitive information, such as login credentials, financial details, or business-related data. Web phishing involves creating fraudulent websites that closely resemble legitimate ones, misleading users into submitting confidential information. Cybercriminals increasingly exploit phishing attacks to conduct security breaches, highlighting the need for more effective detection mechanisms. Existing phishing detection systems lack robust defensive strategies, particularly in mitigating phishing emails and handling limited test data for experimental analysis. This study explores Deep Neural Networks (Deep-NN) as advanced classifiers for optimizing phishing detection performance. A quasi-experimental design is employed, integrating specificity-based filter functions into a Deep Neural Network for unsupervised classification. The research involves building a deep learning model capable of analyzing real-time web traffic data. The dataset, sourced from a web repository, underwent preprocessing to extract relevant features and was cleaned for accurate classification. The experiment utilized 10,000 data phishing instances with 10 selected attributes, resulting in a 10,000 x 10 data dimension for simulation. The implementation was conducted using Python’s machine learning libraries, MATLAB for simulation, and Microsoft Excel for data preparation. The proposed Deep-NN system was evaluated using scientific experiments, achieving an improved phishing detection performance. Key results include a mean square error (MSE) of 1094.6, indicating the average squared difference between classified output and expected target, and a regression (R) value of 0.99, demonstrating a strong correlation between the predicted and actual values. These findings show significant improvement over existing models. This research developed an enhanced model for detecting illegitimate activities in financial web-based applications but further exploration of other unsupervised learning algorithms to filter legitimate emails from large datasets and optimize classification performance is required in the future.
Download Paper