A New Approach for Classifier Model Selection and Tuning Using Logistic Regression and Genetic Algorithms

Abstract

Logistic regression is an efficient machine learning procedure, and it is applied to build a mathematical model for classifying a certain input to a certain class among a number of preset classes. One of the main limitations of the standard classification approaches is the sensitivity to model structure, and another limitation is the sensitivity to the chosen value of regularization parameter λ that affects the estimation of the generalization error of candidate model, as any wrong value might cause underfitting or overfitting. In this research, a new approach for building a classifier model based on logistic regression is proposed, and the new algorithm depends on genetic algorithm to choose the effective model structure and also utilizes a proposed procedure to tune the regularization parameter aiming to find better model parameters. A case study for prediction of a diabetic patient using Pima Indian data set is included, and results showed the high effectiveness of the proposed approach as it reached a simpler model with a higher accuracy.

Time-cost optimization model proposal for construction projects with genetic algorithm and fuzzy logic approach

“…The most recent optimization technique developed in recent years is the genetic algorithm method, which is based on the theory of evolution and the principles of natural selection. The genetic Algorithm used for classification, model selection, and other optimization processes; is the most known metaheuristic optimization method applied to discrete time-cost problems ( Sonmez & Halis, 2012;Aly, 2016).…”