Abstract. The author proposes original tools that are the improvements of formulas in the maximum likelihood estimation method for logistic regression, weight of Eeidence formula, including information value indicator formula, and the Gini coefficient formula to make it possible to use continuous target variable taking on probabilistic values. The research implementation methodologies are the application of the continuous weight functions meeting certain conditions to evaluate the generalized logarithm of the likelihood function, including its generalized gradient vector and generalized Hessian matrix, and application of probability theory to generalize the weight of evidence and the Gini coefficient.
Keywords: logistic regression, weight of evidence, Gini index, maximum likelihood method, credit scoring, reject inference.
Cолошенко Александр Николаевич,
аспирант Национального технического университета Украины «Киевский политехнический институт»,
e-mail: soloshenko_s@ukr.net.