I have tried with no avail to make a logistic regression and random forest model work.

First i fitted a model with climatic measurements as independent variables (2015-2020, almost five years) and mosquito incidence (coded binary for a n regions) data as target. There were some under-sampling issues that were dealt with and the overall models were significant. Some variables were not-significant (or non important for RF) and not kept in the model.

Then, the same (significant) variables were extracted from a future time climate model for 2080-2085, and used to predict the incidence for said region.

The problem is that it for any year on the future time period it predicts presence for all regions.

Anyone has dealt with problems like this? logistic regression or RF models being inconclusive or unusable when used for prediction?

Best