Abstract

Background: Low Birth Weight is a serious public health issue and has major contribution in neonatal morbidity and mortality worldwide. Logistic regression (LR) has been conventionally used to predict low birth weight and identify its risk factors. However, latest data mining techniques like Artificial Neural Network (ANN) have not been used much for this purpose.

Aim: To review the predictive ability of two data mining techniques (Artificial Neural Network and Logistic Regression) for prediction of risk factors of Low Birth Weight.

Methods: All studies that compared predictive ability of ANN and LR for risk factors of LBW were searched on Google scholar, PubMed, Cochran library and web of science using BOOLEAN search strategy and 6 studies following PRISMA guidelines were included. Studies were stored on ENDNOTE version 7 and were critically analyzed. Any disagreements were handled with consensus.

Results: Studies ranged from 1999 to 2019 and all the studies were retrospective cohort. Total of 3,293 subjects were included in all 6 studies. Commonly compared statistical tests were AUC, sensitivity, specificity, negative predictive value, positive predictive value, concordance index, F-statistics, precision and recall. Almost all studies reported that ANN performed better against all these statistical tests or atleast equal in prediction of risk factors of low birth weight.

Conclusion: ANN is a reliable, powerful, and sophisticated tool for handling complex data with high accuracy. ANN can be advantageous over LR specially if considerable inter and intra-relationships of outcome with risk factors and complicated non-linear relationships exist in data.

Keywords: Data mining, Artificial Neural Network, Logistic Regression, Fetal Weight, Low Birth Weight, Pregnancy