| dc.description.abstract | Logistic regression and Naïve Bayes are among the most used data mining classification techniques. However, the performance of the traditional Logistic regression model is weakened by the presence of multicollinearity among the predictor variables which might lead to inaccurate classification results. A number of ways have been suggested to overcome the drawbacks of the traditional Logistic model. These include, the modified Logistic model by incorporating stepwise regression as well as the Naïve Bayes model since it nullifies the assumption of multicollinearity by assuming that the predictor variables are mutually independent, allowing each distribution to be independently estimated. This study intended to compare and evaluate the performance of the traditional Logistic, modified Logistic, Naïve Bayes, and modified Naïve Bayes models in classifying HIV viral load suppression (VLS). The Logistic, modified Logistic, Naïve Bayes, and modified Naïve Bayes models for classifying VLS were developed. The results of the developed models were compared using the 10-fold cross validation (CV) technique on the test dataset. The key metric employed to evaluate the performance of the models was the F-score. The results of the study revealed that the key attributes for classifying VLS includes marital status, HIV clinical stage, ART dosage, adherence, and disclosure. The results also indicated that both the Logistic and modified Logistic models achieved similar F-score results of 79.47%, surpassing the Naïve Bayes and modified Naïve Bayes models attained F-score results of 78.76% and 78.38% respectively. The modified Logistic model demonstrated superior performance compared to the traditional Logistic model on account of utilizing fewer predictor variables but still achieving the same level of classification performance. As a result, the modified Logistic model is considered to be the most optimal model compared to the Logistic, Naïve Bayes, and modified Naïve Bayes models. The stepwise regression approach used by modified Logistic model which eliminated redundant explanatory variables, most likely contributed to improved performance of the model. | en_US |
