Adaptive Feature Engineering for Identifying Spam Tweet Accounts Using Optimal Deep Learning Classifiers

The development of social networks used by people has increased due to technology and the vast amount of data provided to users. Twitter's large user base attracts spammers and illicit activity. To overcome this problem, we propose an Optimized Convolutional Neural Network (OCNN) model to enhance accuracy by developing a reliable and robust spam account detection model on Twitter. Additionally, use the Tweet Cleaning Function to pre-process the data by removing hashtags, usernames, and URLs from the dataset. After that, using the Improved Correlation Feature Selection (ICFS) algorithm, the optimal feature can be selected by removing irrelevant features from the data. Finally, Twitter has proposed an OCNN model based on a Deep Learning (DL) model to provide improved accuracy in classifying accounts as spammers or non-spammers. To detect spam accounts in tweets, finding the best feature should utilize performance parameters such as precision, recall, F1 score, accuracy, and time complexity. The proposed OCNN method achieved an accuracy rate of 96.5% in detecting spam tweet accounts.