Using Machine Learning, Image Processing and Neural Networks to Sense Bullying in K-12 Schools: Enhanced

Abstract

We all have heard about bullying, and we know that it is an immense challenge that schools have to tackle. Many lives have been ruined due to bullying, and the fear it implants into students’ mind has caused many of them to go into depression which can lead to suicide. Traditional methods (National Academies of Sciences, Engineering, and Medicine, in Preventing bullying through science, policy, and practice. The National Academies Press, Washington, DC, [1]) need to be accompanied with modern technology to make the method more effective and efficient. If real-time alerts are sent to school staff, they can identify the perpetuator and extricate the victim swiftly. In this proposed method, an AI-based solution is implemented to monitor students using standard school surveillance technologies and CCTV to maintain a decorum and safe environment in the school premise. Also the proposed method utilizes other unstructured sources such as attendance records, social media activity and general nature of the students to deliver quick response. Artificial intelligence (AI) techniques like convolutional neural networks (CNNs), which include image-processing capabilities, logistic regression methods, long short-term memory (LSTM) and pre-trained model Darknet-19 are used for classification. Further, the model also included sentiment analysis to identify commonly used abuse terms and noisy labels to improve overall model accuracy. The model has been trained and validated with the realistic data from all the sources mentioned and has achieved the classification accuracy of 87% for detecting any sign of bullying.