Cyberbully Detection by Using Machine Learning

Cyberbullying is a growing problem that affects mental health and academic achievement, especially among teenagers and young people. Detecting cyberbullying is challenging due to its complex and subjective nature, as well as the rapidly evolving technology and language. Victims have difficulty avoiding attacks, which can be as simple as malicious comments online. Effective models for detecting cyberbullying are needed to prevent such cases, but data extraction from social media is difficult due to privacy concerns and limited user information. The objective of this project is to enhance manual monitoring for cyberbullying on social networks and online platforms by researching and testing the most effective feature extraction method to be implemented. The project aims to achieve the following goals: preprocess the public dataset, design machine learning algorithm models with different feature extraction methods, evaluate the accuracy and performance of each algorithm, and create a simple interface to test sentences with the designed model. The classifiers used in the research are Support Vector Machine (SVM), Naive Bayes (NB) and Decision Tree (DT) that are optimized with two different feature extraction (TFIDF and Count Vectoriser). The classifier's evaluation is determined by accuracy, precision, recall, and f1-score. The classifier that obtained the highest performance is Linear SVM by using TFIDF.