SUPERVISED LEARNING ON 3D POINT CLOUD DATA: A FEATURE-BASED CLASSIFICATION APPROACH USING HDF5 DATASETS

This study investigates the application of machine learning techniques for demographic prediction from 3D point cloud data, addressing the growing need for privacy-preserving analytics in public space monitoring.The research develops a comprehensive framework that extracts 19 geometric features from human body point clouds, including spatial dimensions, volumetric properties, and morphological characteristics, to predict age categories, gender, and bag-carrying status.Multiple classification algorithms including Random Forest, XGBoost, Support Vector Machine, and Multi-Layer Perceptron were implemented alongside regression techniques such as Linear, Ridge, Lasso, and Polynomial regression for establishing quantitative relationships between geometric features and demographic attributes.The system generates over 75 distinct visualizations encompassing feature distribution analysis, model performance comparison, dimensionality reduction through PCA and t-SNE, clustering patterns, and comprehensive regression analysis.Results demonstrate that height, volume, and body proportions serve as the most predictive characteristics for demographic classification, with Random Forest and XGBoost achieving superior accuracy in cross-validation testing.The framework successfully establishes quantitative relationships between geometric features and demographic variables while maintaining complete privacy protection through anonymous 3D shape analysis.This privacy-conscious approach enables demographic insights for applications in retail analytics, security systems, urban planning, and accessibility research without compromising individual privacy, demonstrating significant potential for real-world deployment in various monitoring and analysis scenarios.