Analysis of CBC and CRP for Children using Machine Learning Models

Abstract

Complete Blood count (CBC) and C- Reactive Protein (CRP) tests provide valuable information about a child's health status and help healthcare providers to diagnose and manage various medical conditions effectively. Paediatric health assessment often relies on these two critical blood tests as normal reports to check the primary condition of the children. Integration of CRP prediction models into clinical decision support systems can facilitate early detection of inflammatory conditions, guide treatment strategies, and improve patient outcomes in paediatric populations. Machine Learning (ML) algorithms are used to enhance diagnostic accuracy, and improve the overall efficiency of paediatric care. The most important test for diagnosing infections is the CBC. The CRP readings are classified into three classes i.e., 0 to 3 mg/L, 3 mg/L to 20 mg/L, more than 20 mg/L. With the test reports of the CBC, prediction of CRP levels was found with highest accuracy from Cat Boosting Classifier [0.976], Neural Network [0.972], Logistic Regression [0.971] and Random Forest Classifier [0.966]. The results prove that the Cat Boosting Classifier performs best among all the used models followed by other algorithms and performs well in predicting the new CRP reading class in all the prediction scenarios given the available dataset. Automating the analysis of CBC and CRP tests can streamline healthcare processes using ML models.