Scholarlink

<p>Statistical packages often provide routines such as nlminb, optim (R), or nlmixed (SAS), which are frequently used to estimate model parameters in statistical research. This paper presents a nature-inspired metaheuristic algorithm called Particle Swarm Optimization (PSO) as a general-purpose optimization tool to solve optimization problems in statistics and highlight its advantages versus conventional algorithms. To this end, we present three cases where PSO is a preferable and more flexible choice to standard packaged routines. In particular, we show that PSO can (1) discover unidentifiable parameters in a complicated model when it is not computationally manifested using routines in R or SAS; (2) circumvent the problem of initial value selection and find admissible estimates for the log-binomial regressions when current routines may not; and (3) accommodate LASSO regularization in log-binomial regressions, which is not readily available in standard packages.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Model Parameters Estimation for the Biosciences Using Particle Swarm Optimization

  • 期刊论文

  • 发表日期：

    2025年2月10日

查看全文
  • Junhyung Park,
  • Sisi Shao,
  • Weng Kee Wong

摘要

Statistical packages often provide routines such as nlminb, optim (R), or nlmixed (SAS), which are frequently used to estimate model parameters in statistical research. This paper presents a nature-inspired metaheuristic algorithm called Particle Swarm Optimization (PSO) as a general-purpose optimization tool to solve optimization problems in statistics and highlight its advantages versus conventional algorithms. To this end, we present three cases where PSO is a preferable and more flexible choice to standard packaged routines. In particular, we show that PSO can (1) discover unidentifiable parameters in a complicated model when it is not computationally manifested using routines in R or SAS; (2) circumvent the problem of initial value selection and find admissible estimates for the log-binomial regressions when current routines may not; and (3) accommodate LASSO regularization in log-binomial regressions, which is not readily available in standard packages.