What action to do after visualizing box plot of variables of a dataset

Before you start to manually select features, I would first try out what the leave-one-out error rate is of a simple classifier like kNN (R function knn from library *class) or a normal distribution model (aka "quadratic discriminant analysis", R functions qda and predict.qda from library MASS). If this does not yoield decent results, you can try to eliminate features, e.g. by greedy backward selection.

Note that your features all seem to cover the same numeric range, but if this is not the case, you might consider standardizing them beforehand to a fixed range (or, alternatively, to varianc one).