Predictive Modeling: Myths and How to Increase the Predictive Power


Key words:  Predictive models, Predictive Modeling, Predictive Power, success and failure, response model, good vs. bad prediction

The following common misperceptions actually reduce the power of your predictive algorithms.

Myth1: The higher the probability of 'good vs. bad' or 'success vs. failure', the better is the discriminating power of the algorithm

Myth2: Positively correlated variable with the 'good vs. bad' or 'success vs. failure' is better than the negatively correlated variable

Myth3: Highest correlated variable is the best variable to improve the prediction  and hence should be used in the model

Myth4: The model building process benefits better if we create binned discrete classes for  independent variables

Myth5: The data is all we have to better predict the consumer behavior