Application of Random Forest Algorithm with Information Gain Feature Selection for Baby Birth Weight Classification

Babies with abnormal birth weights can lead to various serious health issues. Babies with low birth weight are at a higher risk of experiencing stunting, diabetes, hypertension, and heart disease in adulthood. On the other hand, babies with excessive birth weight are at risk of having low blood sugar, jaundice, respiratory problems, and causing delivery complications for the mother. Classifying birth weight is crucial for monitoring baby health and maternal safety. This study aims to classify baby birth weights into three categories: Low Birth Weight, Normal Birth Weight, and Overweight before birth. The dataset used in this study is secondary data derived from the US-BIRTH-2018 dataset. The classification process is conducted using the Random Forest classification algorithm. The use of feature selection methods is necessary to select relevant features. The feature selection process is carried out using the Information Gain algorithm. The research findings conclude that classifying baby birth weights with the Random Forest algorithm without applying the Information Gain feature selection method yields an accuracy, precision, and recall of 92.72%, 89.11%, and $89.08 \%$, respectively. Meanwhile, after applying the Information Gain feature selection method, the highest performance is achieved when using a threshold $\gt$ 0.00001 with 22 selected features, resulting in an accuracy, precision, and recall of $92.81 \%, 89.27 \%$, and $89.22 \%$, respectively. An increase of $0.09 \%$ in accuracy, $0.16 \%$ in precision, and $0.12 \%$ in recall is observed after applying the Information Gain feature selection method, with the three most important features being maternal weight gain, pre-pregnancy maternal weight, and the interval since the previous live birth.