Abstract:
Data mining practices in medical sciences have brought about improved performance in analysis of large and complex datasets. Data mining facilitates evidence-based medical hypotheses. Nowadays, health diseases, especially obstetric fistula, are drastically increasing. According to CCBRT report, approximately 3,000 women suffer from obstetric fistula annually. Since efforts to eradicate obstetric fistula have been inadequate, the researcher was motivated to employ MLA in BIO informatics to detect obstetric fistula in patients. The purpose of this dissertation was to use data mining techniques to predict obstetric fistula. The datasets were collected from CCBRT, Dar es Salaam. In these datasets, there were 367 patient records from January 2015 to February 2019. Out of the 25 attributes, only eight were considered for building the model. These attributes included size of cervix, bigger fetus head, position of the fetus, weight of the fetus, weight of the patients, height of the patients, size of the birth canal, and timing of pregnancy. The environment that was used to evaluate the accurate performance of the predictive model was CV, ROC and CM. The model was trained and tested before being proposed for future use. The research was performed among six different machine-learning algorithms and passed through data cleaning algorithms. The accuracy performance between algorithms prediction shows that LR has better classification accuracies of 87.678%, precision measures of 0.91 (91%), recall measures of 0.82 (82%), f1-score measures of 0.86 (86%) and support measures of 74%. The results of LR in this study are better than those reported in other works of literatures. Thus, the researcher chose to use LR as the proposed obstetric fistula prediction model. LRh as the potential to greatly impact current prediction of obstetric fistula and consequents, treatment, care, and future interventions.
