Using 2019 Crash Data, with the Binary number of Person Collisions, the following method was used to forecast the best algorithm to predict the number of injuries and fatalities in collisions: 
For the Bayes Network: choose the Bayes Classifier, Bayes-Net, with the default K2 search algorithm, set init-As-Naïve-Bayes to TRUE, random-Order to TRUE max-Nr-Of-Parents to 2. run with 10-fold Cross-Validation.  
For the Naïve Bayes: choose the Bayes Classifier, Naïve Bayes, with the default algorithm, set use Supervised-Discretization to TRUE, run with 10-fold Cross-Validation.   
For the J48 decision tree: classify tab, followed by trees, choose J48, with binary splits set to TRUE, reduced error pruning set to TRUE, min-Num-Obj set to 60, with the other options set their default WEKA setting using 10-fold Cross-Validation.    
For K-Nearest Neighbor:
The same WEKA method mentioned above was used for predicting the contributing factor for collisions, with the top 100 collisions over five years (2012-2019), using the training set.