Decision tree based automated prediction of infant low birth weight

Abstract

Data engineering with decision trees plays a vital part in the field of health and medical diagnosis. Low birth weight (LBW) is the single most important factor determining the survival chances of an infant and predicting this LBW based on the maternal factors through data engineering can be another contribution to the medical diagnosis. In this research, decision tree classifiers are used to predict the incidence of low birth weight of newborns based on maternal factors. A set of decision tree classifiers such as C4.5, Random Tree, Random Forest, Decision Stump, Logistic Model Tree, REP Tree (reduced error pruning tree) and BF (best-first) trees are assessed for this classification purpose. These classifiers are evaluated for their accuracy and time complexity in classifying the childbirth weight. A set of data collected from pregnant mothers in the hospitals located in the Eastern part of Sri Lankan have been used for this study. From the experimental analysis, Random Forest produces the highest accuracy as 79.71 % while BF Tree and C4.5 show 79.23%. However, for the time complexity, Random Tree, REP Tree and C4.5 perform less than 1 second whereas the Random Forest utilizes around 5 seconds. On the other hand, BF Tree and C4.5 show precision and recall as 0.792 and 1.0 respectively. Overall, C4.5 could be the best classifier to construct a decision tree model for the prediction of LBW with the acquired data set.