Nonparametric and nonlinear approaches for medical data analysis

• Published in: International journal of data science and analytics Vol. 20; no. 4; pp. 3543 - 3561
• Main Authors: Kakani, Ketana, Radhika, T. S. L.
• Format: Journal Article
• Language: English
• Published: Cham Springer International Publishing 01.10.2025; Springer Nature B.V
• Subjects: Artificial Intelligence; Artificial neural networks; Blood flow; Business Information Systems; Computational Biology/Bioinformatics; Computer Science; Data analysis; Data Mining and Knowledge Discovery; Database Management; Datasets; Decision trees; Effectiveness; Flow characteristics; Machine learning; Performance measurement; Regression analysis; Regular Paper; Nonparametric statistical methods; Kernel estimation; Medical data analysis; Regression splines; Artificial neural networks
• ISSN: 2364-415X; 2364-4168
• DOI: 10.1007/s41060-024-00680-0

Traditional parametric methods are known to impose assumptions about the underlying data distribution, which may not apply to complex medical data. Consequently, nonparametric methods, which require fewer assumptions, and nonlinear models, which can capture complex relationships, are anticipated to provide significant advantages in medical data analysis. This study, therefore, examines the efficacy of these methods in analyzing the medical datasets available to our research team, as our initial investigation revealed that these datasets did not conform to the assumptions required by parametric methods. Accordingly, this work explores the application of nonparametric and nonlinear approaches, reviewing techniques such as kernel smoothing, nearest neighbor algorithms, decision trees, regression splines, and artificial neural networks, and evaluating their efficacy in analyzing a dataset that describes blood flow characteristics in a subject-specific human femoral artery. Our analysis revealed that the regression trees, Nadaraya–Watson kernel estimation, and artificial neural networks outperform other nonparametric methods in predicting blood flow characteristics in this artery. This study also highlights that machine learning models, including parametric approaches, can achieve good performance metrics, such as RMSE and R 2 scores, comparable to nonparametric methods, even when the data deviate from the theoretical assumptions of these models. However, the variance–bias trade-off analysis revealed that nonparametric methods, such as artificial neural networks and kernel estimations, demonstrate their unmatched ability to predict response variables.