I am running some Machine Learning algorithms in order to train a model.

Until now I’ve been doing a correlation matrix in order to select the characteristics with highest correlaction to my target variable.

I read online that doing this selection is not necessary unless I am running Logistic Regression. Is this true?

The algorithms that I am running are Logistic Regression, Decision Tree, SVM, KNN and Naive Bayes.

Should I use my training set with all the characteristics for all the algorithms except Logistic Regression and another version with only the most correlated variables for Logistic Regression?