Application of Neural Networks for Detection of Sexual Harassment in Workspace

Over the several past decades, sexual harassment existed as one of the perturbing issues of society. Sexual harassment at the workplace is even more frightful since the victim might have to go through it on a daily basis without any utterance. Being an unassailable concern sexual harassment at the workplace exists from industry to academia. Not only in developing countries but also in developed countries this objectionable event is increasing in an alarming way. Considering the seriousness of this, a novel approach to reduce such problems has been enlightened in this paper. A Convolutional Neural Network (CNN) based model has been built to detect sexual harassment at workplaces so that it can immediately be verdict or resolved accordingly. Due to the insufficiency of sexual harassment at workplaces dataset, we created our own dataset from social media videos for the model. Conclusively, following transfer learning methods upon three well known pre-trained models of Keras have been implied for getting better accuracy insight of our model. The CNN-based model retained a test accuracy of 95%. On the other hand, the three transfer learning-based models (VGG16, DenseNet201, Xception) possessed 90%, 90%, and 95% test accuracy respectively.