Integral criterion for model training and method of application to targeted marketing optimization

A method of training models to maximize output variable modeling results within a specific interval, comprising:

• providing a target list from a modeling database;
• selecting a model type;
• choosing bounds for the specific interval;
• setting initial weight vectors Wi for the chosen model type;
• starting a first iteration by using the initial weight vectors Wi to calculate output variable classification scores gi for each target from the target list;
• sorting the list by output variable classification score;
• calculating an integral criterion of lift over the range of the specific interval;
• calculating a gradient criterion using the formula:
• Err = ∑ ⁢ ln ⁡ ( 1 + g i ) - ∑ iεresp ⁢ ⁢ ln ⁡ ( g i ) ,
• calculating gradient and non-gradient components of the full error;
• checking the full error for convergence below a tolerance level ε;
• finalizing a new model when convergence occurs;
• and
• calculating new weight vectors Wnew and beginning another iteration by using the new weight vectors to recalculate output variable classification scores for each target from the target list.