Adopting Elastic Net Penalization in Logistic Regression to Achieve Stability: A Case Study of University of UMTHBorno State Nigeria

In this work, study is made on those factors that raise the risk of diabetic patients to be hypertensive and the methods have been determined that best fit the data for modelling. The ordinary logistic regression model fitted turned up to be significant at 1% with a p- value of 0.000. The significant covariates reveal that age increases the risk of being hypertensive for diabetic patient at 16% per unit change. Weight by 60%, marital status decreases it by 74%, exercising decreases it by 95% and family history also by 59%. Various levels of penalization are used for the elastic net logistic regression have shown reasonable improvement over the ordinary logistic regression, with lasso having least deviance ratio (0.52) but also shrinks eight covariates to zero. This is followed by elastic net with centred penalization, which behaves in a similar manner as lasso, but shrinks six of the covariates to zero (with a deviance ratio of 0.58). Ridge logistic regression and the fixed ridge logistic elastic net both retained all the covariates in the model with values of deviance ratio 0.6 and 0.65 respectively.