Scholarlink

In enterprise-scale manufacturing, Material Requirements Planning (MRP) systems play a critical role in aligning supply and demand across complex networks. As organizations expand, MRP executions often face unpredictable runtime and memory usage, which can lead to missed SLAs, infrastructure strain, and delayed planning outputs. This paper presents a data–driven machine learning framework to predict both runtime and memory usage of Oracle–based ASCP MRP plans. The study uses production data from 58 plan runs across 47 organizations and includes extensive feature engineering based on business and system attributes. This study evaluate several regression models including Lasso, Random Forest, Gradient Boosting, and Polynomial Regression with Lasso regularization. The best performing model, Polynomial Lasso, achieved an \(R^{2}\) of 0.998 and RMSE of 1.059 for runtime prediction. SHAP–based analysis was used to improve model transparency and identify key drivers. The results support proactive resource planning, SLA definition, and infrastructure optimization, offering a scalable solution for intelligent forecasting in enterprise planning environments.

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

Machine Learning–Based Prediction of MRP Process Runtime and Memory Usage for Scalable Business Operations

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Bhubalan Mani

摘要

In enterprise-scale manufacturing, Material Requirements Planning (MRP) systems play a critical role in aligning supply and demand across complex networks. As organizations expand, MRP executions often face unpredictable runtime and memory usage, which can lead to missed SLAs, infrastructure strain, and delayed planning outputs. This paper presents a data–driven machine learning framework to predict both runtime and memory usage of Oracle–based ASCP MRP plans. The study uses production data from 58 plan runs across 47 organizations and includes extensive feature engineering based on business and system attributes. This study evaluate several regression models including Lasso, Random Forest, Gradient Boosting, and Polynomial Regression with Lasso regularization. The best performing model, Polynomial Lasso, achieved an \(R^{2}\) of 0.998 and RMSE of 1.059 for runtime prediction. SHAP–based analysis was used to improve model transparency and identify key drivers. The results support proactive resource planning, SLA definition, and infrastructure optimization, offering a scalable solution for intelligent forecasting in enterprise planning environments.