• Àüü
  • ÀüÀÚ/Àü±â
  • Åë½Å
  • ÄÄÇ»ÅÍ
´Ý±â

»çÀÌÆ®¸Ê

Loading..

Please wait....

±¹³» ÇÐȸÁö

Ȩ Ȩ > ¿¬±¸¹®Çå > ±¹³» ÇÐȸÁö > µ¥ÀÌÅͺ£À̽º ¿¬±¸È¸Áö(SIGDB)

µ¥ÀÌÅͺ£À̽º ¿¬±¸È¸Áö(SIGDB)

Current Result Document : 10 / 10

ÇѱÛÁ¦¸ñ(Korean Title) SVM Ä¿³Î ¼±ÅÃÀ» À§ÇÑ Ãßõ ½Ã½ºÅÛ
¿µ¹®Á¦¸ñ(English Title) A Recommender System for Selecting SVM Kernels
ÀúÀÚ(Author) À̱âÈÆ   Ki-Hoon Lee  
¿ø¹®¼ö·Ïó(Citation) VOL 30 NO. 03 PP. 0025 ~ 0034 (2014. 12)
Çѱ۳»¿ë
(Korean Abstract)
Support Vector Machine (SVM)Àº ±¸Á¶Àû À§Çè ÃÖ¼ÒÈ­ ¿ø¸®¸¦ µû¸£´Â Áöµµ ÇнÀ ±â¹ýÀ¸·Î ´Ù¾çÇÑ ºÐ·ù ¹®Á¦¿¡¼­ ³ôÀº Á¤È®µµ¸¦ º¸ÀδÙ. SVM¿¡¼­ Ä¿³Î ÇÔ¼ö´Â ÇнÀ¿¡ °É¸®´Â ½Ã°£°ú Á¤È®µµ¿¡ »ó´çÇÑ ¿µÇâÀ» ¹ÌÄ£´Ù. ±×·¯³ª SVM¿¡ ´ëÇÑ Àü¹®Áö½ÄÀÌ ¾ø´Â »ç¿ëÀÚ´Â ÃÖÀûÀÇ Ä¿³Î ÇÔ¼ö¸¦ ã´Âµ¥ ¾î·Á¿òÀ» °¡Áø´Ù. º» ³í¹®¿¡¼­´Â ¸ÞŸ µ¥ÀÌÅ͸¦ ÀÌ¿ëÇÏ¿© ÁÖ¾îÁø µ¥ÀÌÅÍ ÁýÇÕ¿¡ ÀûÇÕÇÑ Ä¿³Î ÇÔ¼ö¸¦ ÃßõÇØÁÖ´Â ½Ã½ºÅÛÀ» Á¦¾ÈÇÑ´Ù. Á¦¾ÈÇÏ´Â Ãßõ ½Ã½ºÅÛÀÇ µ¿ÀÛ ¹æ½ÄÀº ´ÙÀ½°ú °°´Ù. ¸ÕÀú ÀÎÅͳݿ¡ °ø°³µÇ¾î ÀÖ´Â ´Ù¼öÀÇ SVM ÇнÀ µ¥ÀÌÅÍ ÁýÇÕµéÀ» ¼öÁýÇÏ¿© ´Ù¾çÇÑ Ä¿³Î ÇÔ¼öµé¿¡ ´ëÇØ Á¤È®µµ¸¦ ÃøÁ¤ÇÑ´Ù. °¢ µ¥ÀÌÅÍ ÁýÇÕ¿¡ ´ëÇØ °¡Àå ³ôÀº Á¤È®µµ¸¦ º¸ÀÎ Ä¿³Î ÇÔ¼ö¸¦ ÇØ´ç µ¥ÀÌÅÍ ÁýÇÕÀÇ ¹üÁÖ(class)·Î ¼±Á¤ÇÏ°í µ¥ÀÌÅÍ ÁýÇÕÀÇ ¸ÞŸ µ¥ÀÌÅ͸¦ Ư¡ º¤ÅÍ·Î »ç¿ëÇÏ¿© ÃßõÀ» À§ÇÑ ÇнÀ µ¥ÀÌÅ͸¦ »ý¼ºÇÑ´Ù. »ý¼ºÇÑ ÇнÀ µ¥ÀÌÅ͸¦ ÀÌ¿ëÇÏ¿© ÁÖ¾îÁø »õ·Î¿î µ¥ÀÌÅÍ ÁýÇÕ¿¡ ÀûÇÕÇÑ Ä¿³Î ÇÔ¼ö¸¦ ¿¹ÃøÇÒ ¼ö ÀÖ´Ù. 72°³ÀÇ SVM ÇнÀ µ¥ÀÌÅÍ ÁýÇÕÀ» ÀÌ¿ëÇÏ¿© ½ÇÇèÇÑ °á°ú¿¡ ÀÇÇϸé 90.3%ÀÇ Á¤È®µµ·Î ÃÖÀûÀÇ Ä¿³Î ÇÔ¼ö¸¦ ¿¹ÃøÇÒ ¼ö ÀÖ¾ú´Ù.
¿µ¹®³»¿ë
(English Abstract)
Support Vector Machine (SVM) is a supervised learning technique based on the structural risk minimization principle and shows high accuracy for various classification problems. In SVM, kernel functions have a significant impact on learning time and accuracy. However, users who do not have the expertise on SVM have difficulty in finding the optimal kernel function. In this paper, we propose a system that recommends an appropriate kernel function for a given dataset using metadata. Our recommender system works as follows. It first collects many SVM training datasets publicly available on the Internet and computes accuracy for various kernel functions. The system then generates training data for the recommendation. For each dataset, we assign the kernel function with the highest accuracy as the class of the dataset and use the metadata of the dataset as the feature vector. Using the generated training data, we can predict an appropriate kernel function for a given new dataset. Experimental results using 72 SVM training datasets show that our recommender system predicts the optimal kernel function with 90.3% accuracy.
Å°¿öµå(Keyword) ±â°èÇнÀ   µ¥ÀÌÅÍ ºÐ·ù   SVM   Ä¿³Î ÇÔ¼ö   Ãßõ   machine learning   data classification   SVM   kernel function   recommendation  
ÆÄÀÏ÷ºÎ PDF ´Ù¿î·Îµå