Scholarlink

<p>This paper presents an integrated, scalable Random Forest (SRF)–based predictive framework for estimating the effects of process interventions, including (i) adjusting operating ranges for continuous process parameters within specified tolerances, (ii) selecting specific categories for discrete process parameters, and (iii) combining adjustments to both continuous and discrete parameters. The framework moves beyond linear assumptions by employing a non-linear ensemble approach to identify critical process inputs and quantify their contributions to predicting the process response. These contributions are then leveraged to derive optimal operating ranges for continuous parameters and optimal categories for discrete parameters through a Decision Path Search (DPS) procedure based on tree decision paths. The proposed framework scales to a large number of process factors with complex non-linear dependencies and enables data-driven process improvement. Missing values in mixed-type datasets are addressed using an iterative Random Forest–based imputation scheme, while automatic forest-size optimisation enhances model stability. All preprocessing and modelling steps are embedded within a leakage-safe pipeline, supported by learning-curve analysis and leakage-sanity diagnostics to guard against overfitting. Across the evaluated case studies, SRF delivers accurate predictions together with transparent, practitioner-ready operating windows, translating complex mixed-type manufacturing data into actionable guidance.</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

A scalable random forest (SRF) approach for non-linear predictive modelling using small manufacturing datasets

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年6月1日

查看全文
  • Meshari A. Al-Ebrahim,
  • Rajesh S. Ransing

摘要

This paper presents an integrated, scalable Random Forest (SRF)–based predictive framework for estimating the effects of process interventions, including (i) adjusting operating ranges for continuous process parameters within specified tolerances, (ii) selecting specific categories for discrete process parameters, and (iii) combining adjustments to both continuous and discrete parameters. The framework moves beyond linear assumptions by employing a non-linear ensemble approach to identify critical process inputs and quantify their contributions to predicting the process response. These contributions are then leveraged to derive optimal operating ranges for continuous parameters and optimal categories for discrete parameters through a Decision Path Search (DPS) procedure based on tree decision paths. The proposed framework scales to a large number of process factors with complex non-linear dependencies and enables data-driven process improvement. Missing values in mixed-type datasets are addressed using an iterative Random Forest–based imputation scheme, while automatic forest-size optimisation enhances model stability. All preprocessing and modelling steps are embedded within a leakage-safe pipeline, supported by learning-curve analysis and leakage-sanity diagnostics to guard against overfitting. Across the evaluated case studies, SRF delivers accurate predictions together with transparent, practitioner-ready operating windows, translating complex mixed-type manufacturing data into actionable guidance.