NIPT Temporal Computation Framework Based on Random Forest and Penalty Optimization: Algorithm Design and Efficiency Analysis

View at publisher Summarise with VIRAT
Authors: Xicheng Zhang
Cited by: 0
References: 10

Summary

No summary yet. VIRAT can read this paper and write a plain-language summary of its aim, methods, key findings and limitations.

Summarise with VIRAT Free to try with your Science Catalogs account.

Abstract

This work formulates optimal NIPT scheduling for high-BMI pregnancies as a data-driven decision-optimization task and introduces a hybrid learn–stratify– derivative-free optimization framework: Random Forest regression models cfDNA dynamics with model selection, phenotype stratification is performed via K-means with elbow selection and ANOVA validation, a penalty-based objective aggregates early, late, and out-of-window risks, and simulated annealing yields individualized testing times. Experiments show lower mean squared error and competitive explained variance versus SVR, XGBoost, and MLP under default settings, producing actionable, cluster-specific recommendations with practical deployability. From an algorithmic standpoint, training cost grows with the number of trees, data size, and candidate features per split, inference cost grows with both the number and average depth of trees, clustering cost increases with sample size, number of clusters, and feature dimensionality, and the runtime of derivative-free search is dominated by the number of objective evaluations, each largely governed by model inference. To optimize compute resources, the framework narrows the search space via phenotype stratification to reduce evaluations, applies interval scaling and early stopping to avoid redundant exploration, and exploits parallelism across trees and annealing chains with cached, batched inference to reduce wall-clock time and memory footprint. Overall, the approach improves decision quality under bounded compute budgets and provides a reusable, algorithm-centric blueprint for integrating machine learning into clinical workflows.

Fields of Study & Subjects

Artificial Intelligence in Healthcare Prenatal Screening and Diagnostics Metaheuristic Optimization Algorithms Research

Publication details

Journal / source
IEEE International Conference on Power, Electronics and Green Energy (ICPEGE)
Pages
653–658
Publication date
28 January 2026
Language
EN