Reliability of Predictions Using Hybrid Models: The Case of Malaria Incidence Rates in Uganda

Résumé

Background and purpose: Reliability of estimates emanating from predictive independent data mining techniques is a complex problem. This could be attributed to cross-cutting weaknesses of individual techniques such as collinearity due to high dimensionality of attributes in a dataset, biasedness due to under fitting and over fitting of data as well as noise accumulation due to outliers and thus affecting the reliability of predictions emanating from these models. This study thus aimed at developing a hybrid data mining technique for predicting reliable malaria incidence rate thresholds.

Methods: The decision tree and naïve Bayes classifiers were used to build a hybrid prediction model. Results of the developed hybrid model were compared with independent data mining models using 10-fold cross-validation on a previously unlearned data set. Accuracy, F-measure and the area under the receiver operating characteristics curve (AUC) were the key performance metrics used to evaluate the generalizability of the hybrid model in comparison to the independent models.

Results: Findings revealed that the hybrid classifier attained an accuracy of 79.3% and an F-measure score of 84.2%, the naïve Bayes classifier achieved accuracy and F-measure value of 69% while the decision tree classifier registered an accuracy of 72.4% and an F-measure score of 80%.

Conclusions: The developed hybrid model outperformed both independent decision tree and naïve Bayes models. Notably, the hybrid model outperformed the independent decision tree and naïve Bayes classifiers in terms of accuracy by 6.9% and 10.3% respectively.