Choosing an Appropriate Feature Selection Method to Enhance Feed-Forward ANN

Feature selection (FS) aims to choose a number of attributes from large features subset to have high predictive information and exclude attributes that are irrelevant with least predictive information. This paper utilizes the crash data from Queensland roads available on the Queensland government website. The study's main goal is to identify a subset of attributes from the Queensland roads dataset that can be identified by a feature set based on chosen attributes by applying multiple feature selection methods. ANN Feed Forward classifier used to classify and predict the severity of the injury for the driving accident cases. The results show the effect of multiple feature selection methods, to choose the one which gives high prediction accuracy for ANN classification for driving accidents dataset, by comparing the evaluation results among these FS methods that using a ranked search such as (information Gain, Gain Ratio, Relief-F, and Chi-Squared). The selected features from applying Relif-F results a highest classification accuracy with 80.39% over other FS methods to select ANN data inputs layer.