Scholarlink

Cyber-attacks involve a malicious attack in cyberspace, such as phishing, spamming, and hacking, in order to commit an infraction. Computer systems have been deceitfully compromised and exploited, resulting in massive financial losses that could reach $20 trillion. Email spam is the most well-known form of cyberattack since it consumes the most cyber resources, including processor power, traffic misuse, network bandwidth, etc. Spam emails are unsolicited commercial emails that are sent in large quantities for various purposes. According to studies, almost 85% of modern email is spam. Several tactics against email spamming were presented by individuals, yet they could not give a comprehensive and effective answer. Existing email spam detection strategies have a number of significant shortcomings; first, spam and non-spam emails have not been successfully separated. Second, the time required to detect spam emails increases as the number of received emails grows. Thirdly, the detection filters are put on the server, which causes the server to get overwhelmed when doing massive jobs. Consequently, efficient collaborative detection techniques are needed for detecting and reducing spam emails and their source in a distributed manner at an early stage via effective reaction mechanisms. In this research work, two models are proposed for the prevention of attacks on emails. The first model is responsible for categorizing our mail messages as spammer or non-spammer so that we can only process spam mail for attack prevention. Support vector machines (SVM) are used for label prediction, which gives 98% accuracy and also compares the accuracy with Naive Bayes, which comes out at 88%. Now, for the second model, the data from the first model and the attack dataset are utilized to process the attacks and train them with various classifiers i.e. Logistic Regression, K- nearest neighbour, decision Tree, Extra Tree, Random Forest, Gradient Boosting, Multilayer perceptron (MLP), Gated recurrent unit (GRU), and Long Short-Term Memory (LSTM) and neural networks. Random forest gives 97% accuracy for attack prevention, but when the data is optimized using Particle swarm optimization (PSO) to improve the model's performance, it finally comes out to 98% with Random Forest (RF).

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

A Novel Optimized Classification Approach for Spam Message Cyber Attack Detection and Prevention Using Machine Learning Approach

  • 章节

  • 发表日期：

    2025年

查看全文
  • Sarita Mohanty,
  • Anupa Sinha

摘要

Cyber-attacks involve a malicious attack in cyberspace, such as phishing, spamming, and hacking, in order to commit an infraction. Computer systems have been deceitfully compromised and exploited, resulting in massive financial losses that could reach $20 trillion. Email spam is the most well-known form of cyberattack since it consumes the most cyber resources, including processor power, traffic misuse, network bandwidth, etc. Spam emails are unsolicited commercial emails that are sent in large quantities for various purposes. According to studies, almost 85% of modern email is spam. Several tactics against email spamming were presented by individuals, yet they could not give a comprehensive and effective answer. Existing email spam detection strategies have a number of significant shortcomings; first, spam and non-spam emails have not been successfully separated. Second, the time required to detect spam emails increases as the number of received emails grows. Thirdly, the detection filters are put on the server, which causes the server to get overwhelmed when doing massive jobs. Consequently, efficient collaborative detection techniques are needed for detecting and reducing spam emails and their source in a distributed manner at an early stage via effective reaction mechanisms. In this research work, two models are proposed for the prevention of attacks on emails. The first model is responsible for categorizing our mail messages as spammer or non-spammer so that we can only process spam mail for attack prevention. Support vector machines (SVM) are used for label prediction, which gives 98% accuracy and also compares the accuracy with Naive Bayes, which comes out at 88%. Now, for the second model, the data from the first model and the attack dataset are utilized to process the attacks and train them with various classifiers i.e. Logistic Regression, K- nearest neighbour, decision Tree, Extra Tree, Random Forest, Gradient Boosting, Multilayer perceptron (MLP), Gated recurrent unit (GRU), and Long Short-Term Memory (LSTM) and neural networks. Random forest gives 97% accuracy for attack prevention, but when the data is optimized using Particle swarm optimization (PSO) to improve the model's performance, it finally comes out to 98% with Random Forest (RF).