Crimes Tweet Detection Based on CNN Hyperparameter Optimization Using Snake Optimizer

The increased use of social media has transformed how people share information. In the voluminous social media content, tweets involving criminal activity have increased. The detection and tracking of crime-related tweets helps law enforcement, researchers, and legislators protect the public and identify trends. This study uses Natural Language Processing (NLP) and machine learning (ML) algorithms to identify illicit tweets related to theft, violence, cybercrime, and drug use. Deep convolutional neural networks (CNNs) are used to classify these tweets, but their ideal structure requires further study. The study uses CNN hyperparameters to select relevant features from a dataset, improving model performance. The embedding dimension, dense unit count, learning rate, and batch size are considered. Snake optimization algorithms, inspired by natural processes, are used to provide optimal solutions for complex situations. The paper introduces CNN-SO, a framework that uses a snake optimization algorithm for CNN feature selection of hyperparameter optimization and improving Arabic Twitter crime identification. The model detects crime-related tweets and the results show that CNN-snake optimization outperformed the Conventional CNN used, as the accuracy in Dataset 1 was 99.57% after optimization, and 99.65% with CNN-snake optimization when using Dataset 2, which was built for this model after optimization. However, the accuracy is 88% when using Dataset 1 before optimization CNN, and 87.63% when using Dataset 2, aiming to reduce the burden of these tweets and improve public safety.