Traffic accident is a serious problem worldwide, causing human losses every year. Significant contributors to road accidents are road conditions, climate, unusual driving behaviors, drowsiness, and distraction while driving. In order to mitigate this problem, drivers can be facilitated with a prediction model that can assist them in avoiding accidents. There have been many developments in vehicle crash prediction, but they can be improved in terms of performance and accuracy. This paper suggests an accident prediction model based on Long short-term Neural Networks (LSTM) and Deep Convolution Neural Network (DCNN) Models. The proposed taxonomy allows the creation of a prediction model based on the components such as data, view, and prediction technique. Raw data captured from the gyroscope, speedometer, and smartphone camera is processed for speed estimation. Road facility detection is done through a smartphone-based intelligent Driving Device Recorder (DDR) system consisting of LSTM and CNN. DCNN model is used to analyse different kinds of road components such as traffic lights, crosswalks, stop lines, and pedestrians. Hence, this research critically analyses the works available on vehicle crash prediction using deep learning systems. Furthermore, an enhanced solution that can accurately predict the possible vehicle crash by analyzing the crash dataset using a deep neural network is proposed.