Predictions of Criminal Tendency Through Facial Expression Using Convolutional Neural Network

Criminal intention is a critical aspect of human interaction in the 21st-century digital age where insecurity is on the high side as a major global threat. Kidnapping, killings, molestation of all sorts, gender-based violence, terrorism, and banditry are the trends of criminality in our nation, as such, there is a need to effectively explore innovative means to identify and cope with this evil menace in our society. The facial positioning of humans can tell their evil intention even if they pretended to smile with the evil in their minds. In normal instances, it may be very difficult to predict the heart of man, but with the trending information technology like image processing, the state of a human face could be used as a means to read their tendencies. This paper proposes a deep learning model based on the FER2013 dataset through the implementation of a CNN model that predicts criminal tendencies with the help of facial expressions. With this goal in mind, we explore a new level of image processing to infer criminal tendency from facial images through a convolutional neural network (CNN) deep learning algorithm in other to discriminate between criminal and non-criminal facial images. It was observed that CNN was more consistent in learning to reach its best test accuracy of 90.6%, which contained 8 convolutional layers. To increase the accuracy of this model, several procedures were explored using Random Search from the Keras tuner library, testing out various numbers of convolutional layers and Adam optimizer. It was also noticed that applying the dissection and visualization of the convolutional layers in CNN reveals that the shape of the face, eyebrows, eyeball, pupils, nostrils, and lips are taken advantage of by CNN to classify the images.