Multivariate methods for the analysis of complex and big data in forensic sciences. Application to age estimation in...

Area of Science:

• Data Science
• Forensic Science
• Biometrics

Background:

• High-dimensional datasets present challenges for classical statistical analysis.
• Principal Component Analysis (PCA) has limitations in handling complex data.
• Forensic sciences increasingly face large, high-dimensional datasets.

Purpose of the Study:

• To introduce and evaluate novel nonlinear dimensionality reduction techniques (NLDR).
• To compare NLDR with PCA for analyzing complex datasets.
• To explore the application of NLDR in age estimation for living persons.

Main Methods:

• Application of NLDR to a theoretical dataset.
• Comparison of NLDR and PCA results.
• Application of NLDR to clinical, dental, and radiological data (13 variables) for age estimation.
• Estimation of data correlation dimension.

Main Results:

• NLDR techniques outperformed PCA.
• NLDR revealed significant informational redundancy (correlation dimension of 2).
• The study highlighted potential discrepancies in age estimation even for individuals with similar characteristics.

Conclusions:

• NLDR techniques are recommended for analyzing complex and big data, potentially outperforming or replacing PCA.
• Current data used for age estimation may require re-evaluation for suitability.
• Further research is needed on integrating diverse data and approaches to enhance age estimation accuracy.