Comparison of simple and stratified random sampling on porn videos recognition using CNN

Video classification is challenging because the video consists of many frames. In the videos recognition system, the proper sampling method affects the classification process because it uses image recognition model on each frame to recognize the video. This study focuses on comparing the sampling methods used in pornographic video recognition systems. Porn videos have high heterogeneity so that a sophisticated approach is needed to analyze the provided data for learning the data patterns. The Convolutional Neural Network method is employed because it can automatically detect features or similarity from the given training data. Besides that, this method could recognize the image quickly because it just fit test data into the final weights and biases from training. In this research, the stratified random sampling method gives 80% of accuracy while the simple random sampling method is the fastest method, which recognizes the video in 94 seconds. Additionally, all porn videos provided can be identified entirely so that the recall value for all test video is 100% while specificity average is 55.2%.