Machine Learning Comparative Analysis for Enhanced Identification Potential of Clinical Features from Medical Data

This paper explores the application of some well-known machine learning (ML) algorithms for efficient identifying insights from medical data. The study focuses on three specific algorithms: AdaBoost, XGBoost and k-Nearest Neighbor (k-NN), in a comparative evaluation of their performance. The investigation was conducted using data obtained from the Smoker's Health Data database, which includes more than 3,900 records with variables such as age, sex, heart rate, blood pressure and smoking status. The performance of each algorithm was evaluated based on accuracy and training/evaluation time. The results indicated that XGBoost achieved the highest accuracy (0.88) for the proposed task, followed by AdaBoost (0.85) and k-NN (0.82). However, k-NN was the fastest in terms of training and evaluation time. Performed analysis shows the potential of ML algorithms in medical diagnosis, especially in the context of personalized healthcare and predictive analytics. The study highlights the strengths but also the limitations of each algorithm. Future research could focus on further optimizing these algorithms and exploring their use in other medical conditions.