Machine Learning Prediction for Appendicitis: A Comprehensive Study of Blood Markers and Clinical Variables

Appendicitis is an illness frequently seen in young people. It is an acute condition that often manifests within a few hours and necessitates emergency surgery. The symptoms manifest within 24 hours and, in some cases, gradually (over weeks or months). The accurate diagnosis of appendicitis is the most important safety measure in preventing dangerous, needless appendectomy. Blood markers and clinical variables play a vital role by providing quantifiable data that can ensure the accuracy of diagnosis and the execution of emergency decisions. We employ machine learning (ML) models to analyze blood markers and clinical variables to predict appendicitis, aiming to enhance diagnostic accuracy. These ML models: Support Vector Machines (SVM), Random Forests (RF), Gaussian Naive Bayes (GNB), AdaBoost with Decision Trees (ABDT), and Multilayer Perceptron (MLP), using a dataset downloaded from Figshare and published by Erdoan, Alirza in 2021 with a digital object identifier (DOI) as https://doi.org/10.6084/m9.figshare.16908478. An extensive evaluation of blood markers and clinical variables was performed, and models were trained (75 %) and tested (25 %) to predict appendicitis. The result shows that ABDT outperformed all other models on a wide range of parameters, including 0.97 for accuracy, precision, recall, and F-1 score, 0.98 for Area Under the Receiver Operating Characteristic Curve (AUC-ROC), and 0.96 for Area Under the Precision-Recall Curve (PR-AUC). The statistical comparison of ABDT and SVM is significant with a p-value of 0.0335. This study highlights the usefulness of ML models for appendicitis prediction, their potential for early diagnosis, better patient outcomes, and more efficient healthcare delivery.