Enhancing Predictive Modeling in High Dimensional Data Using Hybrid Feature Selection

Healthcare data has become invaluable for patients care enhancement as well as outcomes which have been influenced by the expanding volume and complexity of health informatics. The current research discusses the design of a hybrid feature selection(HFS) methodology which is based on a combination of filter & wrapper methods to detect the necessary variables for predictive modeling using machine learning algorithms. The study takes the data set on Parkinson's disease as a use case. The performance of KNN, Decision Trees, and Random Forest models is assessed based on this data set. Findings reveal that integration of feature selection impacts models' efficacy, where Random Forest performs best among all techniques. KNN helped to lower the total number of errors but to increase sensitivity, while the Decision Trees (DT) applied changes not only to sensitivity but to specificity also. The findings bring feature selection into the limelight making healthcare data analysis more revealing in disease prediction and personalized medicine. The application of filter and wrapper methods by providers makes the data exploring process in a high-dimensional data easier and useful in getting actionable insights for improved patient care. Additional studies on feature selection methods and network architecture evolution should be carried out in order to increase the efficiency of medical data mining and enhance predictive model accuracy.