Scholarlink

This study investigates OCR-driven information extraction and ensemble learning for product type classification from Thai receipt data to enhance family expense management. Using 1,305 receipt images from Thailand, we extracted and preprocessed 5,087 product names across five categories. We compared base classification algorithms with ensemble learning algorithms, focusing on their performance in handling OCR-extracted Thai text. Results demonstrated the superiority of ensemble methods, particularly Majority Voting and Extra Trees, in classifying product types. Majority Voting achieved a weighted average F1-score of 91.74% and accuracy of 91.92%, while Extra Trees recorded the highest overall accuracy at 92.05%. This study contributes to the field by addressing the unique challenges of Thai language OCR and product classification, offering insights into effective ensemble learning strategies for receipt data analysis.

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

Leveraging OCR-Driven Information Extraction for Accurate Product Type Classification from Thai Receipt Data: An Ensemble Learning Approach

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • Wongpanya S. Nuankaew,
  • Apitarat Autarach,
  • Teerapakorn Meesri,
  • Pratya Nuankaew

摘要

This study investigates OCR-driven information extraction and ensemble learning for product type classification from Thai receipt data to enhance family expense management. Using 1,305 receipt images from Thailand, we extracted and preprocessed 5,087 product names across five categories. We compared base classification algorithms with ensemble learning algorithms, focusing on their performance in handling OCR-extracted Thai text. Results demonstrated the superiority of ensemble methods, particularly Majority Voting and Extra Trees, in classifying product types. Majority Voting achieved a weighted average F1-score of 91.74% and accuracy of 91.92%, while Extra Trees recorded the highest overall accuracy at 92.05%. This study contributes to the field by addressing the unique challenges of Thai language OCR and product classification, offering insights into effective ensemble learning strategies for receipt data analysis.