Regression and neural networks analysis in vesco-vaginal fistula causality: a comparative approach

Abstract

Vesico vaginal fistula (WF) is an abnormal opening of the vaginal wall to the bladder or rectum resulting in the leakage of urine. It is one of the worst morbidities associate with delivery and is a major public health problem on the rise with an estimated minimum of 150,000-200,000 patients in Nigeria. Neural network are able to solve the nonlinear regression problem. Very little research has been conducted to model the causes of WF using artificial neural networks. The data set obtained from the case records of women admitted with cases of Vesico-vaginal Fistula (WF) in Maryam Abacha Women and Children Hospital Sokoto, from January 2000 to December 2010 was used. We then compared the performance of Statistical neural networks and Regression model. In comparison to traditional methods, the value of Obstructed labour and misuse of instrument in ANN has higher R square (0.8 & 0.54) in which is a better result, lower MSE (2011 &4S79.6) which is also a better result. The p-value is only greater than 0.05 in obstructed labour. The results of the t and F statistics confirms the better performance, since any p-value lesser than 0.05 shows that that cause of WF cases is very significant. Therefore, we can accept the fact that MISUSE OF INSTRUMENT and YANKAN GISHIRI are both significant to cases of WF using ANN, while LR is not since the R squares are low. Statistical neural network model showed better predictions than various regression models for causes of WF. However, both methods can be used for the prediction of causes of WF.