In current situation, video classification is one of the important research domains. Since video is a complex media with various components, classification of video is normally a complex process. This paper presented a human action based movie trailer classification using optimized deep convolutional neural network in video sequences. Initially, images are converted into the grayscale conversion. Using the adaptive median filtering process, the pre-processing stage is accomplished. Threshold based segmentation approach is utilized for subtracting the background from the video frames and to extract the foreground portion. In the feature extraction stage, the visual features (color and texture features) and motion features are extracted from the segmented portion. Finally, the mined features are trained and classified with the help of optimized deep convolutional neural network (DCNN) for the movie trailer classifications. Here, the deer hunting optimization (DHO) is introduced to optimize the weight values of DCNN. The proposed (DCNN-DHO) human action based movie trailer classification is executed in the MATLAB environment. The experimental results are evaluated and compared with the existing methods in terms of accuracy, false alarm rate, sensitivity, specificity, precision, F-measure and false discovery rate. The results of the proposed method are compared with filtering process and without filtering process in which 95.23% of accuracy is achieved for the suggested approach with filtering and 90.91% of accuracy is achieved for the suggested approach without filtering process.