Extreme Learning Machine

Slow speed of feedforward neural networks has been hampering their growth for past decades. Unlike traditional algorithms extreme learning machine (ELM) [5][6] for single hidden layer feedforward network (SLFN) chooses input weight and hidden biases randomly and determines the output weight through linear algebraic manipulations. We propose ELM as an auto associative neural network (AANN) and implement it as a single class classifier and ensemble the results (AAELF). The proposed architecture consists of three layers namely the input layer, hidden layer and the output layer which is same as the input layer. The proposed algorithm has been tested on bankruptcy prediction datasets namely Spanish banks, Turkish banks, UK banks; Credit UK dataset and phishing dataset. AAELF achieved higher sensitivity for phishing dataset when compared with algorithms used by Mayank and Ravi[1] and Credit UK and UK banks yielded better results for AAELF compared to PSOAANN [2]. It is concluded that ELM can be modified as an AANN and is an effective tool in classifying datasets, where the samples are disproportionately distributed between positive class and negative class such as phishing mails, bankruptcy datasets.