Abstract: Multi-target prediction is a prediction that consider more than one target variable in a real-life problems like cervical cancer simultaneously instead of the concentration of most researchers on supervised learning that has to do with prediction of a single target variable. The framework for multiple target variables has significant effect for categorization and evaluation that a single target variable framework cannot take care of. In the findings in the course of this study we did not come across the use of multi-target regression technique for predictive performance measure on cervical cancer dataset that predict all the target variables simultaneously. In this study, we adopt the problem transformation approach using multi-target classifiers to transform a binary classification task into a regression task. The predictive performance measures in supervised learning for multi-target classification task employ in this study is evaluated using exact match, hamming loss, hamming score, ZeroOne loss and accuracy per label. The findings of this study shows that the multi-target classifier (Bayesian classifier chains) using decision stump (base classifier) gives the highest predictive performance measures on hamming score, exact match, ZeroOne loss and accuracy per label compared to the multi-target classifier (classifier chains and class relevant) using J48 and random forest (base classifier) using 10 folds cross-validation and training and testing evaluation option. In conclusion, this study support the assertion made by some researchers that decision tree and random forest are powerful techniques for prediction.
S.G. Fashoto, A.S. Metfula, B.B. Matsebula and B.Y. Fashoto, 2018. Multi-Target Regression Prediction on Cervical Cancer for Evaluation of Predictive Performance Measures. Asian Journal of Information Technology, 17: 160-166.