Inappropriate Video Detection for Kids Through AI

Abstract

The internet hosts a variety of videos for children, including cartoons and other types of media. Some of these videos may be inappropriate for kids, containing harmful content that promotes violence or aggressive behavior, which can negatively impact a child’s mental health. Moreover, YouTube’s recommendation algorithm often suggests content that appears similar to appropriate material, making it challenging to differentiate between suitable and unsuitable videos. This study aims to address this issue by applying Machine Learning (ML) and Convolutional Neural Networks (CNN) techniques to detect inappropriate videos. The research compares the accuracy of different models, showing that the Random Forest Classifier achieves the highest accuracy (0.999) for ML, while the Xception model performs best among deep learning methods.