In this study, a weakly-supervised video anomaly detection approach using modified anomaly score module and modified BERT is proposed. The proposed approach consists of four main steps. (1) A video sequence is divided into nonoverlapping video clips and each video clip consists of sixteen adjacent video frames. (2) The video clips are fed into I3D to extract video clip feature vectors, which are fed into the proposed modified anomaly score module and the proposed modified BERT module to obtain the video clip feature vector anomaly scores and the video sequence classification anomaly score, respectively. (3) The video clip feature vector anomaly scores are multiplied by the video sequence classification score to obtain the video clip anomaly scores, which are fed into frame anomaly score production operation to obtain the video frame anomaly scores. (4) Based on the specified threshold, the video frames are determined as abnormal or normal. Based on the experimental results obtained in this study, the performance of the proposed approach is better than those of comparison approaches.