I have consolidated my research skills by authoring and publishing a research paper titled ‘Prediction System Design for Monitoring the Health of Developing Infants from Cardiotocography Using Statistical Machine Learning’ in the Design Engineering, a Scopus International Journal.
Today segregation models are widely used in health care, which is intended to support physicians in diagnosing diseases and reducing human error. The challenge is to use effective methods to extract real-world data from the medical field, as many different models have been proposed with varying results. Many researchers have focused on the problem of variability in real-time data sets in segmentation models. Some previous works create mechanisms that include similar graphs for information display and information acquisition. However, such methods are weak in finding different relationships between elements. The purpose of this diagnostic method is to measure the baby's heart rate and uterine contractions during the third trimester of pregnancy, when the baby's heart is fully functional. Cardiotocogram findings are usually divided into three categories: physical, suspicious, or pathological. The purpose of this work is to automatically distinguish these regions using cardiotocographic data.
Cardiotocography is a low-cost means of monitoring fetal health and a method used for lowering the child fatality rate. Visual analysis errors were one of the most serious problems with CTG monitoring. Any type of interventional surgery, whether necessary or not, raises the risk of complications. My ML model effectively differentiated normal and disturbed fetal health and was accurate enough in its predictions to avoid any unwanted interventive surgical procedures to reduce the overall risk of child mortality. In the entire research paper, I have done machine learning classification on fetal health dataset, starting with exploratory data analysis where we found the pattern and distribution of each feature in the dataset using different types of a graph such as a heatmap, histogram, and data distribution, etc. The problem of class imbalance was resolved by using SMOTE (Synthetic oversampling minority techniques) and ADASYN (Adaptive Synthetic Sampling Approach) are data sampling techniques. A well-trained model helped in identifying which variable changes to the FHR had the most significant effect on fetal health. I used five different types of classification algorithms like Logistic Regression, Random Forest, Decision Tree, Support Vector Machine, and Naive Bayes classifier. These algorithms were applied on each of the three different sampled data which consisted of Actual, SMOTE, and ADASYN data. The statistical metrics I used to differentiate the five algorithms were accuracy, precision, recall, and f1-score. On comparing the results, the best performing algorithm was Random Forest Classifier on SMOTE data sample. It gave the highest accuracy of 94% among all the remaining classifiers. Further, I used Grid Search CV and found out the best model parameters for the best metrics.