Web-phishing attacks are one of the most serious cybercrime. It enables hackers to access the devices of many users and spy on their personal data such as passwords and credit card details. Hackers use a lot of tricks through the internet, which make users to share data, download files or open links that attack a computer. This research proposes meta-heuristic based approach to protect the internet users from the web-phishing. It consists of three phases, the first phase uses a new proposed method for evaluating and ranking the features of URL, HTML and JavaScript code, text, images and domain name of the web page. The second phase extracts the effective subset of the ranked features that achieves the highest classification accuracy of the web-phishing. The third phase constructs the Random forest classifier training by data features of the extracted subset. The new proposed method of the feature selection achieved the highest classification accuracy compared to the correlation feature selection, information gain, principle component analysis, and Relief feature selection algorithms. The proposed methodology of the web-phishing detection was also evaluated, it obtained the highest classification accuracy at the least possible time compared to the adaptive Neuro-fuzzy inference system.