Phishing attacks are always surfacing as key threats against internet users, necessitating advanced detection methods. Blacklist-based systems and rule-based models of phishing detection generally have had critical limitations in dealing with evolving tactics and new phishing schemes. Some of these approaches fail to cope with the temporal and visual patterns of phishing sites, which are crucial for timely and accurate detection. To overcome these difficulties, this work introduces a hybrid AI-based phishing website detection model that utilizes several machine learning and deep learning techniques to improve the accuracy of the detection and remove false positives. The proposed model uses LSTM networks, Genetic Algorithms, Random Forest, and CNN through the stacking ensemble framework. Since LSTM is adopted to capture the temporal dependencies in the website traffic and user interaction patterns, this model can effectively model their phishing behavior over time. GA is used for bioinspired feature selection to reduce the dimensionality of features while optimizing model performance. Random Forest is used as a base layer addressing structured features like URL characteristics and WHOIS information. CNNs are incorporated to extract feature content from a webpage and images that carry various visual indicators often used in phishing attacks including counterfeit logos or banners. A meta-classifier is then used to combine the outputs of LSTMs, CNN, and RF and generate the final classification to boost the detection rate. The proposed hybrid model surpasses the existing techniques and facilitates the analysis of temporal, visual, and structured data, making the detection considerably more accurate. Achieving accuracy of as much as 96-97% and having an AUC of 0.97 with a false positive rate below 3%, the model then impacts the more powerful and more flexible phishing detection system, which is then capable of being more protective against higher sophisticated phishing techniques with increased precision and fewer false alarms.
Design of a Hybrid AI- based Phishing Website Detection using LSTM, CNN, and Random Forest based Ensemble Learning Analysis
2024-11-06
569623 byte
Conference paper
Electronic Resource
English
A Hybrid LSTM-Based Ensemble Learning Approach for China Coastal Bulk Coal Freight Index Prediction
DOAJ | 2021
|Fault Diagnosis of MVB Based on Random Forest and Ensemble Pruning
British Library Conference Proceedings | 2020
|Flood Hazard Rating Prediction for Urban Areas Using Random Forest and LSTM
Springer Verlag | 2020
|