เอกสารการประชุมวิชาการและเสนอผลงานวิจัย มหาวิทยาลัยทักษิณ ครั้งที่ 19 2552 - page 188

2
„µ¦ž¦³»¤ª·µ„µ¦Â¨³Áœ°Ÿ¨Šµœª· ‹´¥ ¤®µª·š¥µ¨´¥š´„¬·– ‡¦´ÊŠš¸É 19 ž¦³‹Î µže 2552
¦³—´‡nµŸ·—ž„˜· ´—nªœ„µ¦ž¨°¤žœ ¨³ œµ—˜´ª°¥nµŠ ¤¸Ÿ¨˜n°‡nµ MAPE °Šª·›¸„µ¦ž¦³¤µ–‡nµ
´¤ž¦³·š›·Í„µ¦™—™°¥š´Ê Š 3 ª·›¸
‡n µ MAPE °Šš»„ª· ›¸¤¸ÂœªÃœo¤¨—¨Š Á¤ºÉ ° œµ—˜´ª°¥nµŠÁ¡·É¤ ¹Ê œ –³š¸É ´—nªœ„µ¦ž¨°¤žœ‡Šš¸É
‡n µ MAPE °Šš»„ª· ›¸¤¸ÂœªÃœo¤Á¡·É ¤ ¹ÊœÁ¤ºÉ°´—nªœ„µ¦ž¨°¤žœÁ¡·É¤ ¹Êœ –³š¸Éœµ—˜´ª°¥nµŠ‡Šš¸É
‡ÎµÎµ‡´: ¤„µ¦™—™°¥Ã¨‹· ˜·‡ ˜´ ªž¦³¤µ–‡ªµ¤‡ª¦‹³Áž}œ¼Š» — ˜´ªž¦³¤µ–š¸É ¤¸‡ªµ¤Â„¦nŠ ¦o°¥¨³ °Š‡nµ
‡¨µ—Á‡¨ºÉ°œÁŒ¨¸É¥´¤¼¦–r
Abstract
The objective of this research is to compare the estimation methods of parameters in logistic regression
model with outliers . The estimation methods are Weight Maximum Likelihood Method of Croux and Haesbroeck,
Weight Maximum Likelihood Method of Rousseeuw and Christmann and Weight Maximum Likelihood Method of
Daniel of parameter. There are two level of outliers, mild and extreme, and four proportions of contamination 0.00,
0.05, 0.10 and 0.15. Mean Absolute Percentage Error (MAPE) is the criterion of comparison. The sample sizes are 20,
50 and 100. Simulation data and Monte Carlo method are used to compute MAPE. The experiment was repeated until
the difference of parameters between of iteration k and 1 k less than 0.0001 under each situations. Results of the
research are as follows:
The smallest MAPE is Weight Maximum Likelihood Method of Daniel and Weight Maximum Likelihood
Method of Rousseeuw and Christmann and Weight Maximum Likelihood Method of Croux and Haesbroeck
repectively, when level of outliers and proportion of outliers contamination increase. Whereas sample sizes more than
50 , the MAPE of all methods are nearly the same.
MAPE of parameters increase when level of outliers and proportion of outliers contamination increase but
they decrease when sample sizes increase.
The level of outliers, proportion of outliers contamination and sample sizes have effected on parameters
estimates.
Keywords : Logistic regression; Maximum likelihood estimator; Robust estimator; Mean Absolute Percentage Error
‡ÎµœÎµ
ž{ ‹‹»´œÅ—o¤¸„µ¦œÎµ‡ªµ¤¦¼ošµŠ—oµœ™· ˜·Åžž¦³¥»„˜r čo„´Šµœ—oµœ˜nµŠÇ ×¥ÁŒ¡µ³°¥n µŠ¥·ÉŠ„µ¦«¹„¬µ¦¼ žÂ
‡ªµ¤´¤¡´œ›r ¦³®ªnµŠ˜´ªÂž¦š¸ÉÁ¦¸¥„ªnµ¦¼žÂ„µ¦™—™°¥ Ž¹É ŠÁž}œ„µ¦ª· Á‡¦µ³®ro°¤¼¨Á¡ºÉ°°›·µ¥™¹ Š‡ªµ¤´¤¡´œ›r¦³®ªnµŠ
˜´ªÂž¦˜µ¤š¸ÉÁž} œÁ·Šž¦· ¤µ–¨³˜´ªÂž¦°· ¦³š¸ÉÁž} œ˜´ªÂž¦Á·Šž¦·¤µ–°¥nµŠÁ—¸ ¥ª ®¦º°°µ‹¤¸µŠ˜´ªÂž¦°· ¦³š¸ÉÁž}œ˜´ª
ž¦Á·Š‡»–£µ¡ ®¦º°Á· Š„¨»n¤ ˜n¤¸ŠµœµŠž¦³Á£šš¸É¡ªnµ˜´ ªÂž¦˜µ¤Áž}œ˜´ªÂž¦Á· Š‡»–£µ¡®¦º°Á· Š„¨»n¤š¸É¤¸‡nµÁž} œÅž