Effective large for gestational age prediction using machine learning techniques with monitoring biochemical indicators

摘要

A newborn with a birth weight above the 90th percentile of same gestational age is termed as large for gestational age. Large for gestational age suffers from serious complications during and after the antepartum period because they do not get earlier identification of the disease. Earlier recognition of large for gestational age infants could slow progression and prevent further complication of the disease. In medical science, prevention and mitigation of disease require examination of biochemical indicators. Machine learning has been evolved and envisioned as a tool to predict large for gestational age infants with most deterministic characteristics. This study aims to identify most deterministic biochemical indicators for large for gestational age prediction with minimal computational overhead. To the best of my knowledge, this is the first time a study is carried out to identify the most deterministic risk factors associated with large for gestational age and to develop large for gestational age prediction model using machine learning techniques. To develop an efficient large for gestational age prediction model, we conducted three group of experiments that considered basic machine learning methods; feature selection; and imbalanced data, respectively. Support vector machine, logistic regression, Naive Bayes and Random Forest were trained using tenfold cross-validation on large for gestational age dataset; we selected precision and area under the curve as a performance evaluation metrics; information gain an entropy-based feature selection method was adopted to rank features; we introduced an ensemble data imbalance technique in the last group of experiments. For each group of experiments, support vector machine performed best compared to other machine learning classifiers by producing the highest prediction precision score of 85%. All of the classifiers performed best with thirty ranked features subset, which validates the applied method to recognize the most deterministic risk factors associated with large for gestational age prediction.