Scholarlink

<p>Heteroscedasticity is a problem in regression, where data exhibits different levels of variance across observations. Techniques like data transformations, weighted least squares, Huber-White standard errors have been used to address this problem. This paper evaluates the effectiveness of two target encoding schemes (Box-Cox and Yeo Johnson transformation) in addressing heteroscedasticity, which has not been explored exhaustively in the previous work. Our research combines these transformations with Kernelized Extreme Learning Machine(KELM) to evaluate various regression datasets to demonstrate the model’s efficiency in handling heteroscedasticity. These target encoding schemes are also integrated with the You Only Look Once version 5(YOLOv5) model to evaluate their impact on improving the accuracy and robustness of Traffic Sign Detection (TSD), where heteroscedasticity may occur due to change in shape, size or appearance of an object in different regions or conditions within an image. The root mean square error (RMSE) is used as an evaluation metric, where after applying transformations led to a reduction in RMSE ranging from 65% to 95% in regression datasets, and up to 90% reduction in the detection of traffic signs, highlighting the efficiency of our proposed work.</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

Handling heteroscedasticity in kernelized extreme learning machine based regression and deep learning models for traffic sign detection using box cox and Yeo Johnson transformations

  • 期刊论文

  • 发表日期：

    2025年9月5日

查看全文
  • Manali Chandnani,
  • Sanyam Shukla,
  • Rajesh Wadhvani

摘要

Heteroscedasticity is a problem in regression, where data exhibits different levels of variance across observations. Techniques like data transformations, weighted least squares, Huber-White standard errors have been used to address this problem. This paper evaluates the effectiveness of two target encoding schemes (Box-Cox and Yeo Johnson transformation) in addressing heteroscedasticity, which has not been explored exhaustively in the previous work. Our research combines these transformations with Kernelized Extreme Learning Machine(KELM) to evaluate various regression datasets to demonstrate the model’s efficiency in handling heteroscedasticity. These target encoding schemes are also integrated with the You Only Look Once version 5(YOLOv5) model to evaluate their impact on improving the accuracy and robustness of Traffic Sign Detection (TSD), where heteroscedasticity may occur due to change in shape, size or appearance of an object in different regions or conditions within an image. The root mean square error (RMSE) is used as an evaluation metric, where after applying transformations led to a reduction in RMSE ranging from 65% to 95% in regression datasets, and up to 90% reduction in the detection of traffic signs, highlighting the efficiency of our proposed work.