mds2$REPORT_NUMBER<- NULL
mds2$REPORT_DATE <- NULL
mds2$VEHICLE_ID_NUMBER  <- NULL

df<- mds2 %>% rename(tag = FATALITIES)

data2<-ohse(df)

b=data2[sample(nrow(data2),replace=F,size=0.20*nrow(data2)),]

# Using random forest for variable selection
rfModel <-randomForest(tag~ ., data = b,do.trace=TRUE,importance = TRUE)

# Getting the list of important variables
# make dataframe from importance() output
feat_imp_df <- important_variables(measure_importance(rfModel), k = 10)

varImpPlot(rfModel)
#######################################################

h2o.init()

data3<-cbind(data2[,feat_imp_df],data2$tag)


results <- h2o_automl(data3, y="data2$tag",
                      project = "Arthur",
                      max_time = 30,
                      seed = 123,plots = TRUE)
plot(results$plots$dashboard)

results$plots$importance