Cardiotocogram Data classification using Machine Learning Algorithms along with Swarm-Based Metaheuristic optimizations for Autonomous Fetal distress detection.

Abstract

Fetal distress is a critical condition that requires timely and accurate detection to warrant the safety of birth mother and fetus. Uterine contractions of birth mother and foetus heart rate are generally monitored by a non-invasive technique called Cardiotography. However, the manual interpretation of CTG data is often subjective and prone to errors. This study presents the autonomous detection of fetal distress using various metahuriestic optimization techniques for machine learning algorithms. The dataset of CTG recordings from UCI repository is utilized and pre-processed to apply for various machine learning algorithms including Naïve Bayes (NB), k-Nearest Neighbour (k-NN), Support Vector Machines (SVM) and Random Forest (RF) that are employed to classify the CTG data into various categories such as normal, suspect, and pathological. To further improve the classification accuracy, we integrate nature-inspired metaheuristic optimization methods such as Firefly, Grasshopper and Greywolf algorithms to fine tune the hyperparameters select appropriate features. Greywolf optimized model with Random Forest classifier algorithm outperformed other models with overall accuracy of around 93.65%, Weighted F1 score of 94.12% with Mean MCC and Mean Kappa score of about 82.86% and 81.70% respectively and an average ROC-AUC of 92.78% suggests that Greywolf optimized model with Random Forest algorithm is a reliable tool for the early detection of fetal distress. The results indicate that the integration of machine learning with metaheuristic optimization not only enhances the predictive capabilities but also provides a robust framework for autonomous fetal monitoring systems.