how (logistic regression, random forest) deal with input zero values? I am new in ML. In my dataset there are 11 of 21 features that have some zero values.
what is the impact of having zero values as input when using logistic regression or random forest to train my model?
 A: You can have zero in your features, that's not a problem unless your ML algorithm is very very specific/custom, which is not in this case. Logistic regression multiples your sample's features with weights, i.e. $x_i$'s with $w_i$'s, and makes a decision based on the result; and if $x_i$ is $0$ for a sample, it's just $i$-th branch is not activated in the neuron for that specific sample. Here, don't assume that $x_i=0$ means $i$-th branch doesn't have a saying on the decision just because it is multiplied with $0$. For the random forest, or decision trees to put it simply, having $0$ has absolutely no effect. You can add 1 to all your samples and obtain the same decision tree again since it just sorts the feature values and finds a threshold.
A: One of the requirements for a logistic regression is that the independent variables have little or no multicollinearity between them (i.e. should not be too highly correlated with each other). So, a potential risk is if a substantial portion of the values in your variables are zeroes, this could increase the correlation between them.
