Online Program Home
  My Program

All Times EDT

Abstract Details

Activity Number: 60 - Invited E-Poster Session II
Type: Invited
Date/Time: Sunday, August 8, 2021 : 6:45 PM to 7:30 PM
Sponsor: Section on Statistical Learning and Data Science
Abstract #317399
Title: Using Machine Learning Algorithms with Bayesian Optimization Turning Technique for Fracture Prediction in Genomic and Phenotypic Data of 25,772 Postmenopausal Women
Author(s): Qing Wu* and Jingyuan Dai
Companies: University of Nevada, Las Vegas and University of Nevada, Las Vegas
Keywords: Machine Learning ; Bayesian optimization; Data sciences ; Genomics; prediction; Major osteoporotic fracture
Abstract:

Bayesian Optimization tuning has the potential to enhance the parameter-tuning process in machine learning(ML). We aimed to determine if the Bayesian Optimization technique could improve the ML model’s performance in fracture prediction compared to the traditional grid search method. Genomic and phenotypic data of the Women’s Health Initiative (N=25,772) was used. We identified 1,103 risk SNPs and derived the genetic risk score(GRS) from the data. GRS and clinical risk factors were used to predict major osteoporotic fracture in the follow-up. The data were normalized and randomly split into a training set (70%) and a validation set (30%). Random forest, gradient boosting, neural network, and other ML models were built and tuned using the Bayesian optimization technique and the grid search method separately. The 10-fold cross-validation was used for model tuning in the training set. ML model’s prediction performance was assessed using the testing set. We found that each ML model with Bayesian optimization tuning had higher accuracy and higher AUC than using grid search. Thus, the Bayesian optimization technique improves ML performance in fracture prediction in postmenopausal women.


Authors who are presenting talks have a * after their name.

Back to the full JSM 2021 program