Phishing websites are amongst the biggest threats Internet users face today, and existing methods like blacklisting, using SSL certificates, etc. often fail to keep up with the increasing number of threats. This paper aims to utilise different properties of a website URL, and use a machine learning model to classify websites as phishing and non-phishing. These properties include the IP address length, the authenticity of the HTTPs request being sent by the website, usage of pop-up windows to enter data, Server Form Handler status, etc. A Support Vector Machine binary classifier trained on an existing dataset has been used to predict if a website was a legitimate website or not, by finding an optimum hyperplane to separate the two categories. This optimum hyperplane is found with the help of four optimization algorithms, the Bat Algorithm, the Firefly Algorithm, the Grey Wolf Optimiser algorithm and the Whale Optimization Algorithm, which are inspired by various natural phenomena. Amongst the four nature-inspired optimization algorithms, it has been determined that the Grey Wolf Optimiser algorithm’s performance is significantly better than that of the Firefly Algorithm, but there is no significant difference while comparing the performance of any other pair of algorithms. However, all four nature-inspired optimization algorithms perform significantly better than the grid-search optimized Random Forest classifier model described in earlier research.

仿冒网站是当今互联网用户面临的最大威胁之一，现有的方法（如将其列入黑名单，使用SSL证书等）通常无法跟上不断增长的威胁。本文旨在利用网站URL的不同属性，并使用机器学习模型将网站分为网络钓鱼和非网络钓鱼。这些属性包括IP地址长度，网站发送的HTTPs请求的真实性，使用弹出窗口输入数据，服务器表单处理程序状态等。已经对现有数据集进行了训练的支持向量机二进制分类器。通过找到将这两个类别分开的最佳超平面来预测网站是否为合法网站。这个最佳超平面是通过四种优化算法（蝙蝠算法）找到的，受各种自然现象启发的萤火虫算法，灰狼优化器算法和鲸鱼优化算法。在四种受自然启发的优化算法中，已确定Gray Wolf Optimiser算法的性能明显优于Firefly算法，但是在比较任何其他算法对的性能时都没有显着差异。但是，所有四种受自然启发的优化算法的性能都明显优于早期研究中描述的网格搜索优化的随机森林分类器模型。已经确定，Gray Wolf Optimiser算法的性能明显优于Firefly算法，但是在比较任何其他算法对的性能时，都没有显着差异。但是，所有四种受自然启发的优化算法的性能都明显优于早期研究中描述的网格搜索优化的随机森林分类器模型。已经确定，Gray Wolf Optimiser算法的性能明显优于Firefly算法，但是在比较任何其他算法对的性能时，都没有显着差异。但是，所有四种受自然启发的优化算法的性能都明显优于早期研究中描述的网格搜索优化的随机森林分类器模型。