Phishing attacks have become more sophisticated in web-based transactions. As a result, various solutions have been developed to tackle the problem. Such solutions including feature-based and blacklist-based approaches applying machine learning algorithms. However, there is still a lack of accuracy and real-time solution. Most machine learning algorithms are parameter driven, but the parameters are difficult to tune to a desirable output. In line with Jiang and Ma’s findings, this study presents a parameter tuning framework, using Neuron-fuzzy system with comprehensive features in order to maximize systems performance. The neuron-fuzzy system was chosen because it has ability to generate fuzzy rules by given features and to learn new features. Extensive experiments were conducted, using different feature-sets, two cross-validation methods, a hybrid method and different parameters and achieved 98.4% accuracy. Our results demonstrated a high performance compared to other results in the field. As a contribution, we introduced a novel parameter tuning framework based on a neuron-fuzzy with six feature-sets and identified different numbers of membership functions different number of epochs, different sizes of feature-sets on a single platform. Parameter tuning based on neuron-fuzzy system with comprehensive features can enhance system performance in real-time. The outcome will provide guidance to the researchers who are using similar techniques in the field. It will decrease difficulties and increase confidence in the process of tuning parameters on a given problem.
|Number of pages||11|
|Publication status||Published - 9 Oct 2014|
|Event||2014 Science and Information Conference (SAI) - London, UK|
Duration: 27 Aug 2014 → 29 Aug 2014
|Conference||2014 Science and Information Conference (SAI)|
|Period||27/08/14 → 29/08/14|