Applying Extreme Learning Machine with Structure Tuning for a Phishing Websites Classification

The rapid growth of data in many applications of intelligent systems created a need for a proper and timely analysis to optimize the performance of machine learning models. Researchers are applying different techniques to minimize the time required to train machine learning models that are capable of performing the required tasks efficiently. Extreme Learning Machine (ELM) is known for its fast training time and good generalization performance. For this purpose, this paper proposes weight neural network (AU-ELM) based on automatic construction of ELM, by selecting the appropriate model hyper-parameters that are convenient with the phishing website type, to get a better classification for every observed website. The proposed method is applied and tested on a phishing website classification problem. To evaluate the performance of the proposed AU-ELM, it has been compared with a models created based on a set of rules cited from the literature to determine the number of hidden neurons required in SLFN. The results showed that the AU-ELM enhances generalization performance and outperforms the other static methods.