Road Accident Prediction and Feature Analysis By Using Deep Learning

With the growing number of road accidents in the city Kerala, it is authoritative to carry out road accident mitigation processes to prevent damage of many lives. Handling of information mining has usually encountered issues like noise, high dimensionality and imbalance data. High dimensionality denotes to the large quantity of attribute among the data set. On the off chance that the quantity of attributes in a datasets is huge, at that point it is in all likelihood that a large portion of these qualities are repetitive or futile for building an prediction model and better result can be accomplished if the redundant attributes from the dataset are removed. Feature selection techniques are used to select the subset of the features which is used in the learning process. The SMOTE technique is used to handle the imbalance of the Kerala road data set then δ - agree based boosting algorithm for stacked autoencoder techniques are used for feature selection as well as data classification analysis. From this research, we conclude about various causes of the accidents, then find out total accidents per annual in separate state in particular time interval and helps to find accident trend in each state in each time interval over the year in each state.