Scholarlink

<p>This study presents an innovative approach for detecting and estimating outliers in time series data, emphasizing constrained-remaining components decomposition. The method extends the moving linear model to accommodate outliers, resulting in an enhanced moving linear model. A state-space representation improves computational efficiency through Bayesian estimation. We introduce a novel method for determining outlier positions, starting with initial estimates of the remaining components. The proposed methodology combines maximum likelihood and Bayesian-type estimation for effective outlier detection and estimation, guided by the minimum Akaike Information Criterion (AIC). Furthermore, we investigate outlier detection in time series data with seasonal components. Applications to real data, specifically the Index of Industrial Production (IIP) and Wholesale Commercial Sales (WCS) in Japan, showcase the simplicity and potential for automation in the proposed approach, making it a promising tool for time series analysis, particularly in constrained-remaining components decomposition.</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

Enhancing business cycle analysis by integrating anomaly detection and components decomposition of time series data

  • 期刊论文

  • 发表日期：

    2025年1月16日

查看全文
  • Koki Kyo

摘要

This study presents an innovative approach for detecting and estimating outliers in time series data, emphasizing constrained-remaining components decomposition. The method extends the moving linear model to accommodate outliers, resulting in an enhanced moving linear model. A state-space representation improves computational efficiency through Bayesian estimation. We introduce a novel method for determining outlier positions, starting with initial estimates of the remaining components. The proposed methodology combines maximum likelihood and Bayesian-type estimation for effective outlier detection and estimation, guided by the minimum Akaike Information Criterion (AIC). Furthermore, we investigate outlier detection in time series data with seasonal components. Applications to real data, specifically the Index of Industrial Production (IIP) and Wholesale Commercial Sales (WCS) in Japan, showcase the simplicity and potential for automation in the proposed approach, making it a promising tool for time series analysis, particularly in constrained-remaining components decomposition.