以支持向量機界定農地重金屬高污染潛勢區

୯ҥᆵ᡼εᏢғၭᏢଣғނᕉნس಍πำᏢس ᅺγፕЎ

Department of Bioenvironmental Systems Engineering College of Bioresources and Agriculture

National Taiwan University Master Thesis

аЍ࡭ӛໆᐒࣚۓၭӦख़ߎឦଯԦࢉወ༈୔

Delineation of Heavy Metal Pollution Potential Areas in Agricultural Land Using Support Vector Machine

Analysis

ቅঅሎ Xiu-Ming Liu

ࡰᏤ௲௤Ǻ஭൧୯ റγ Advisor: Tsun-Kuo Chang, Ph.D.

ύ๮҇୯ 104 ԃ 6 Д

ᄔा

Ѝ࡭ӛໆᐒȐSupport Vector Machine, SVMȑឦܭၗૻࣽᏢሦୱύޑᐒᏔᏢಞ ȐMachine learningȑǴࣁ΋ᅿᅱ࿎ԄᏢಞȐSupervised LearningȑᄽᆉݤǴځϩᜪǵ

ӣᘜޑфૈҭёᔈҔܭӦ፦ǵᕉნࣽᏢ฻࣬ᜢሦୱǶҁࣴزஒၭӦख़ߎឦ֖ໆ

ፓࢗၗ਑೸ၸϣఘᛥࡰ኱ȐNemerow index, PNȑᙯඤࡕǴа SVM མଛӦ౛ၗ

ૻس಍Ȑgeographic information system, GISȑჄϩβᝆၭӦख़ߎឦଯԦࢉወ༈୔Ǵ ၸำύ೸ၸ10 ԛҬΰᡍ᛾ᓬᒧ૽ግ໣኱ᠸಔԋКٯǵኬҁኧໆǶ่݀ᡉҢǴᄆϯ ᑜа7,353 ฽ᗺՏӧ໚܄ȐPNʁ1.0ȑǵ഍܄ȐPN<1.0ȑ኱ᠸК 1Ǻ2 Πࡌҥϐኳࠠ

຾Չβᝆख़ߎឦԦࢉወ༈ႣෳǴ่݀ྗዴࡋȐAccuracyȑࣁ 85.37%ǵF1-measure ࣁ0.692ǹਲ༜ѱӧ኱ᠸКࣁ 1Ǻ1 ΠǴӅ 3,288 ฽ၗ਑ኳࠠǴԦࢉወ༈Ⴃෳϐ่

݀ྗዴࡋࣁ 71.58 %ǵF1-measure ࣁ 0.506Ƕ٠ஒ่݀঺᠄ݞοࢬୱǵπቷǵπ཰

୔฻ޜ໔ϩѲၗૻǴຑ݋аSVM ჄϩၭӦख़ߎឦԦࢉወ༈୔ୱϐᆬӢϷᜢೱ܄Ǵ

᛾ჴ SVM ᄽᆉݤૈԖਏӦᔈҔܭβᝆख़ߎឦԦࢉወ༈ჄϩǴЪӧե૽ግ໣ኬҁ ኧջёၲؼӳޑϩᜪਏૈǶ

ᜢᗖຒǺЍ࡭ӛໆᐒǵख़ߎឦǵβᝆԦࢉǵԦࢉወ༈კ

Abstract

Support Vector Machine (SVM) is a kind of supervised learning algorithm of machine learning in computer science, it’s function such as classification and regression could also be applied to related field e.g. geoscience and environmental science. In this research, the data of heavy metal pollution areas in agricultural land converted by Nemerow index (PN) combined with SVM and geographic information system (GIS) classifies the highly potential heavy metal pollution areas in agricultural land. For modeling, the samples were optimized into an ideal proportion for training data set by using 10-fold cross validation. In Changhua County, at 7,353 points with the sample labeled ratio of positive (PNɪ1.0) and negative (PN<1.0) set to 0.5, results show the potential heavy metal pollution area with an accuracy of 85.37% and F1-measure of 0.692; In Taoyuan city, at 3,288 points with sample labeled ratio set to 1, results show the potential heavy metal pollution area with accuracy of 71.58% and F1-measure of 0.506. By interpreting the mapping of results with the information of surrounding geological features such as the distribution of river basins, factories and industrial zones, it allowed us to divide the causes and relationships of potential heavy metal polluted area with the use of SVM. Thus, the algorithm had proved that it could be validly applied to classify the potential heavy metal pollution areas in agricultural land even with low training data set.

Keywords: Support Vector Machine, heavy metals, soil pollution, pollution potential map

Ҟᒵ

ᄔा ... I Abstract ... II Ҟᒵ ... III კҞᒵ ... V ߄Ҟᒵ ... VI

ಃ΋ക ᆣፕ ... 1

1.1 ࣴز୏ᐒ ... 1

1.2 ࣴزҞޑ ... 2

1.3 ࣴزࢎᄬ ... 2

ಃΒക Ў᝘ӣ៝ ... 4

2.1 Ѝ࡭ӛໆᐒȐSupport Vector Machine, SVMȑว৖ϷᔈҔ ... 4

2.1.1 ਡЈڄኧȐkernel functionȑϷୖኧᒧۓ ... 6

2.1.2 ᆛ਱ཛྷ൨Ȑgrid-searchȑϷҬΰᡍ᛾ȐCross Validationȑ ... 7

2.2 షౄંତȐconfusion matrixȑ ... 8

2.3 ϣఘᛥࡰ኱ȐNemerow Index, PNȑᆶϩભ ... 9

ಃΟക ࣴز׷਑ǵБݤϷࢬำ ... 11

3.1 ࣴز׷਑ ... 11

3.1.1 ࣴز୔ୱ ... 11

3.1.2 ၭӦፓࢗၗ਑ ... 14

3.2 ࣴزБݤ ... 15

3.2.1 ϣఘᛥࡰ኱ȐPNȑ ... 15

3.2.2 Ѝ࡭ӛໆᐒȐSVMȑ ... 17

3.3.2 ᆛ਱ཛྷ൨ ... 20

ಃѤക ่݀ᆶ૸ፕ ... 21

4.1 ϣఘᛥࡰ኱ȐPNȑຑ݋่݀ ... 21

4.1.1 ૽ግ໣ኬҁၗ਑኱ᠸ ... 24

4.2 ૽ግ໣኱ᠸಔԋКٯ௖૸ ... 26

4.3 ૽ግ໣ኬҁኧໆ่݀௖૸ ... 27

4.4 ၭӦख़ߎឦԦࢉᜢᖄ܄ຑ݋ ... 42

ಃϖക ่ፕ ... 49

ୖԵЎ᝘ ... 50

კҞᒵ

კ 1-1 ࣴزࢎᄬკ ... 3

კ 3-1 ᄆϯᑜՉࡹ୔ୱ ... 12

კ 3-2 ਲ༜ѱՉࡹ୔ୱ ... 13

კ 3-3 Ѝ࡭ӛໆᐒཷۺҢཀ ... 17

კ 3-4 നε໔႖ຬѳय़ ... 18

კ 3-5 Ѝ࡭ӛໆᐒȐSVMȑϩᜪࢬำ ... 20

კ 3-6 ᆛ਱ཛྷ൨ࢬำ ... 20

კ 4-1 ᄆϯᑜӚ௦ኬᗺ PNϩѲკ ... 22

კ 4-2 ਲ༜ѱӚ௦ኬᗺ PNϩѲკ ... 23

კ 4-3 ᄆϯᑜሚॶࣁ 1.0 ϐӚ௦ኬᗺ PN ... 24

კ 4-4 ਲ༜ѱሚॶࣁ 1.0 ϐӚ௦ኬᗺ PN ... 25

კ 4-5 ᄆϯᑜόӕ૽ግ໣ኧໆᆶᆛ਱ཛྷ൨่݀ ... 29

კ 4-6 ᄆϯᑜόӕ૽ግ໣ኧໆᆶଯԦࢉወ༈ϩѲࣚۓ ... 32

კ 4-7 ਲ༜ѱόӕ૽ግ໣ኧໆᆶᆛ਱ཛྷ൨่݀ ... 36

კ 4-8 ਲ༜ѱόӕ૽ግ໣ኧໆᆶଯԦࢉወ༈ϩѲࣚۓ ... 39

კ 4-9 ᄆϯᑜଯԦࢉወ༈୔ᆕӝࣴ݋კ ... 43

კ 4-10 ਲ༜ѱଯԦࢉወ༈୔ᆕӝࣴ݋კ ... 44

კ 4-11 ᄆϯᑜНճλಔຑ݋ԦࢉϷӒ্฻ભ঺᠄ SVM ޑႣෳ่݀ ... 45

კ 4-12 ਲ༜ѱНճλಔຑ݋ԦࢉϷӒ্฻ભ঺᠄ SVM ޑႣෳ่݀ ... 46

კ 4-13 ᄆϯᑜಃ III ભӥၡڀԦࢉወ༈୔ୱ঺᠄ SVM ޑႣෳ่݀ ... 47

კ 4-14 ਲ༜ѱಃ III ભӥၡڀԦࢉወ༈୔ୱ঺᠄ SVM ޑႣෳ่݀... 48

߄Ҟᒵ

߄ 2-1 షౄંତȐConfusion matrixȑ ... 8

߄ 2-2 PNӦ߄Нϐϩભ኱ྗ ... 10

߄ 2-3 PNӦΠНϐϩભ኱ྗ ... 10

߄ 2-4 PNβᝆၭӦख़ߎឦϩભ኱ྗ ... 10

߄ 3-1 ࣴز୔ཷॊ ... 11

߄ 3-2 ᄆϯᑜၭӦ߄βख़ߎឦᐚࡋፓࢗ಍ी ... 14

߄ 3-3 ਲ༜ѱၭӦ߄βख़ߎឦᐚࡋፓࢗ಍ी ... 14

߄ 3-4 PNޑຑ݋୷ྗॶȐCsiȑ ... 15

߄ 3-5 ၭӦख़ߎឦϩભ኱ྗ ... 16

߄ 4-1 ᄆϯᑜ PNϩભϩѲ߄ ... 22

߄ 4-2 ਲ༜ѱ PNϩભϩѲ߄ ... 23

߄ 4-3 ᄆϯᑜ PNϩભȐሚॶ೛ۓࣁ1.0ȑ ... 24

߄ 4-4 ਲ༜ѱ PNϩભȐሚॶ೛ۓࣁ1.0ȑ ... 25

߄ 4-5 ᄆϯᑜኬҁКٯჴᡍ่݀ ... 26

߄ 4-6 ਲ༜ѱኬҁКٯჴᡍ่݀ ... 26

߄ 4-7 ᄆϯᑜόӕ૽ግ໣ኧໆჹᔈႣෳ่݀ ... 28

߄ 4-8 ਲ༜ѱόӕ૽ግ໣ኧໆჹᔈႣෳ่݀ ... 35

߄ 4-9 Չ཰ᜪձϷځख़ߎឦԦࢉᅿᜪ ... 42

ಃ΋കġ ᆣፕ

1.1 ࣴز୏ᐒ

βᝆࢂΓᜪᒘағӸޑ୷ҁԾฅၗྍǴҭࢂၭ཰ว৖ޑख़ा୷ᘵǶၸѐኧΜ ԃ໔Ǵᆵ᡼π୘཰ଯೲว৖ǴβᝆௗڙٰԾឲ෸ǵ௝শǵక࿼฻೼৩ޑಕᑈǴԋ ࣁӚᅿԦࢉނϐനಖ܍ڙᡏǶ

ၭӦβᝆԦࢉޑٰྍǴЬाҗНԦࢉᒿ๱ឲ෸෤ၰ܈ឲ௨෤ၰЇΕၭӦǴ೷

ԋख़ߎឦӧၭӦύූ੮ಕᑈǶԶᅿ෌ଯ֖ໆख़ߎឦϐၭӦޑНዿǴځख़ߎឦ֖ໆ ҭڀԖଯԦࢉϐወ༈ǴځύаၭӦᙿǵႉԦࢉࣁЬǴϦ্٣ҹ߾೷ԋᙿԯ٣ҹࣁ നӭȐ஭Ǵ2002ȑǶ 1950 ԃܭВҁ൤ξᑜวғӄౚಃ΋ଆᙿԯԦࢉ٣ҹǴፓࢗځ চӢࢂ྽Ӧ᝜཰௨ܫрख़ߎឦᙿԿݞούǴၭ҇ЇឲډၭҖаठᅿрᙿԯǴ٠೷

ԋ၀Ӧۚ҇ᎁڙภภੰӒ্Ƕ 1982 ԃܭਲ༜ᑜᢀॣໂε዇׸วғಃ΋ےᆵ᡼ᙿ ԯ٣ҹǴӢ྽Ӧϯπቷғౢ֖ᙿکႉޑӼۓᏊǴᒿ෤ၰ௨ܫԿၭӦ܌ठǹᒿࡕᄆ ϯᑜǵѠύᑜǵ໦݅ᑜ฻ᑜѱҭௗೱวғᙿԯԦࢉ٣ҹǴЇว୯҇ᜢݙǶਥᏵၭ

ہ཮಍ीǴ߈Μԃٰᆵ᡼؂Γ؂ԃқԯឪڗໆѳ֡ऊ 49 ϦАǴᙿԯୢᚒ໪཈ख़

ೀ౛ȐChen & Lee, 1995; Hsu et al., 2010; Yao et al., 2014ȑǶ

ԐයᕉߥൂՏࢂ௦س಍܄௦ኬБԄǴ२Ӄஒᆵ᡼аᆛ਱Ⴤϩࡕ຾ՉၭӦख़ߎ ឦԦࢉፓࢗǴӆ٩Ᏽፓ่ࢗ݀ஒख़ߎឦԦࢉၨᝄख़ϐ୔ୱ຾Չᆛ਱ಒϩ٠ፓࢗǴ а೴؁ว᝺ԦࢉጄൎǶҗܭβᝆ௦ኬϷᅱෳԋҁܳ຦ǴЪЖલԖਏ౗ϸࢀଯԦࢉ

ወ༈୔ୱϐБݤǶӧ౜жࣽמ຾؁ΠǴཛྷ൨Ԧࢉ዗ᗺϐᄽᆉݤ࡭ុว৖Ǵ೸ၸஒ ၸ ۳ ၭ Ӧ β ᝆ Ѯ ໆ ፓ ࢗ ၗ ਑ Ǵ ೸ ၸ ᑔ ᒧ ᐒ ڋ ǵ Ӧ ౛ ၗ ૻ س ಍ ȐGeographic Information System, GISȑǵᅱ࿎ԄᏢಞ฻᏾ӝϩ݋ǴаԖਏБݤዴᇡଯԦࢉወ༈

ϐၭӦ୔ୱǴפрख़ߎឦଯԦࢉወ༈ϐβᝆ୔ୱǴගଯβᝆԦࢉᅱෳǵٛݯ฻ਏ

౗Ǵှ،ׯ๓ᆵ᡼Ӧ୔βᝆԦࢉୢᚒǶ

1.2 ࣴزҞޑ

ҁࣴزϐҞޑࣁᔈҔᐒᏔᏢಞǴаЍ࡭ӛໆᐒȐSupport Vector Machine, SVMȑ

୔ϩၭӦख़ߎឦଯԦࢉወ༈୔ǶచӈҞޑӵΠǺ

1. ᔈҔЍ࡭ӛໆᐒஒၭ၂܌ޑ௦ኬၗ਑Ǵुۓሚॶǵ኱ᠸϷ૽ግኳࠠǴ٠೸ၸႣ

ෳϩᜪǵᡍ᛾ǴᑔᒧрଯԦࢉወ༈୔ୱǶ

2. ೸ၸፓ᏾૽ግၗ਑໣Ȑtraining data setȑޑόӕ኱ᠸКٯϷኧໆǴ௖૸ځჹ SVM ϩᜪ่݀ϐቹៜำࡋǶ

3. ஒ SVM ϩᜪ่݀঺᠄᏾ӝᕉნӦ߄੝ቻǴຑ݋βᝆख़ߎឦଯԦࢉወ༈୔ୱϐ ԦࢉᆬӢǶ

1.3 ࣴزࢎᄬ

ҁࣴزϐࢎᄬӵკ 1-1 ܌ҢǶ

! ! !

კ 1-1 ࣴزࢎᄬკ

ಃΒകġ Ў᝘ӣ៝

2.1 Ѝ࡭ӛໆᐒȐSupport Vector Machine, SVMȑว৖ϷᔈҔ

Ѝ࡭ӛໆᐒȐSupport Vector Machine, SVMȑࣁ΋ᅿਥᏵ಍ीᏢಞ౛ፕ ȐStatistical Learning TheoryȑԶว৖рޑᐒᏔᏢಞБݤǴനԐҗ Bernhard E. Boser

฻ӧ1992 ԃޑीᆉᐒᏢಞ౛ፕȐComputational Learning Theory, COLTȑޑࣴ૸཮

ගрཷۺǶCorinna Cortes ک Vladimir N. Vapnik ӧ 1995 ԃȠMachine Learningȡ යтύว߄ȐBoser et al., 1992; Cortes & Vapnik, 1995ȑǶӢځᓬذЪڀଯྗዴ܄ޑ ϩᜪૈΚǴЪૈೀ౛ଯᆢࡋޑၗ਑ǴSVM ౜Ϟς೏ቶݱӦၮҔӧೀ౛ϩᜪޑୢᚒ

΢Ƕ

SVM ܭӦౚࣽᏢᆶᕉნሦୱϐᔈҔԖǴ1999 ԃܭВϣґ෫ȐLake GenevaȑǴ

೸ၸ෫ۭ؈ᑈނளډᙿȐCdȑϐᐚࡋϩѲǴ٠аሚॶȐthreshold valueȑࣁ 0.8 Ɋgڄg^-1Ϸ 1.0 Ɋgڄg^-1ஒ૽ግၗ਑୔ϩǴаޜ໔০኱ǵᙿᐚࡋၗ਑೸ၸ SVM ࡌҥ ኳࠠǶќѦ٬Ҕෳ၂໣բྗዴࡋᡍ᛾Ǵ٠຾΋؁Ⴃෳ᏾ঁ෫ۭޑख़ߎឦᙿᐚࡋϩ ѲȐMikhail et al., 1999ȑǶ

2008 ԃܭύ୯Ѥο࣪ϐؙοεӦ᎜ǴаෟԢࢬୱӦ᎜Їวྖڵǵଯำǵڵفǵ ڵࡋ฻ၗ਑ǴբࣁSVM ૽ግၗ਑ϐᜪձǶ่݀ளډྖڵ௵ག܄ࢀ৔ϐ୔ୱϩѲǴ

ࣴز૸ፕύаόӕྖڵኧໆբࣁ૽ግၗ਑Ǵ௖૸ჹ่݀ྗዴࡋϐቹៜȐChong et al., 2012ȑǶ

2010 ԃܭ༞ᅟᆢ٥Ǵ೸ၸβᝆፓࢗޑۓໆמೌٰႣෳکϩᜪβᝆᅿᜪǶځύ

೸ၸ௦ኬᗺϣ֖ϐpHǵNǵK2OǵP2O5ǵCECǵclay sandȐℚβࣳȑ฻բࣁ૽ግ ϐᜪձǴҔаࡌҥኳࠠ٠Ⴃෳ่݀Ǵஒβᝆ୔ϩࣁ໵້βǵరྋβ฻ᅿᜪȐMiloš et al., 2010ȑǶ

2012 ԃܭҲਟլᅟୗ࣪ȐKerman provinceȑޑ Now Chun ᝜׉୔ǴAbedi

᝜ౢ୎ࢗǶԜࣴز่݀ᡉҢǴЍ࡭ӛໆᐒёբࣁ᝜ނ߻ඳޑႣෳǴ٠Ԗਏගଯ୎

௖߻ඳወ༈୔ୱޑϩᒣ౗کफ़եᢕϔ॥ᓀȐMaysam et al., 2012ȑǶ

2012 ԃܭՋ੤УޑЉఘ܎৞ȐGomera IslandȑǴ၀Ӧ౛୔Տ္Ǵϩձа໺಍

ၭ཰ǵԖᐒၭ཰ݤᅿ෌ؽ݀Ƕ೸ၸβᝆύԖ٤ߎឦ֖ໆǴӵ້ǵႃǵልǵ៓ǵႇǵ ᗔǵᒰǵ໊ǵᙻکᎋޑᐚࡋ֖ໆǴբࣁᜢ߯ᜪձ٠аSVM ૽ግϩᜪǶ่݀ᡉҢ ଯၲ93 %ႣෳૈΚǴྗዴࡋ࣬྽ଯȐHernᙻndez-Sᙻnchez et al., 2012ȑǶ

җ΢ॊЎ᝘ளޕǴSVM ܌ੋϷޑᕉნ࣬ᜢቫय़ߚதቶݱǴவӦ౛ᕉნǵβᝆ

܄፦ǵ௖୎ፓࢗǵۭݝख़ߎឦᐚࡋϩѲ฻ሦୱࣣԖ఼ᇂǴЪ่݀ᡉҢӧ୔ϩᜪձ

΢ڀԖؼӳਏૈǶ

2.1.1 ਡЈڄኧȐkernel functionȑϷୖኧᒧۓ

SVM ё೸ၸᒧ᏷ਡЈڄኧȐkernel functionȑೀ౛ᔈҔߚጕ܄ୢᚒȐZuo et al., 2011ȑǴаڄኧφࢀ৔ډଯᆢࡋޜ໔Ǵ؃૽ግӛໆ໣x_iϐጕ܄ന٫ຬѳय़ȐOptimal

HyperplaneȑǶۓကK(x_i,x_j)≡φ(x_i)^T(x_j)^T ᆀࣁਡЈڄኧǴதـԖаΠѤᅿǺ

1. linearǺK(x_i,x_j)= x_i^Tx_j

2. polynomial kernelǺK(x_i,x_j)=(γ x_i^Tx_j +r)^d, γ ∈ℜ⁺

3. radial basis functionȐRBFȑǺK(x_i,x_j)=exp (−γ x_i −x_j ²)^, γ ∈ℜ⁺

4. sigmoidǺK(x_i,x_j)= tanh(γ x_i^Tx_j +r) ځύǴ

γ

^{, r, dࣣࢂਡЈୖኧǶ}

җܭጕ܄ਡЈ Ȑlinear kernelȑࢂ RBF ύޑ੝ٯȐKeerthi & Lin, 2003ȑǹsigmoid ਡЈӧ፾྽૽ግୖኧΠǴϩᜪՉࣁႽRBFȐLin & Lin, 2003ȑǶԜѦǴᆶ RBF ࣬ КǴӭ໨ԄਡЈȐpolynomial kernelȑሡाፓ᏾ၨӭୖኧǴᏤठࡌҥኳࠠǵᒧ᏷ୖ

ኧਔၨܭፄᚇǴЪӧ Huang ฻Ȑ2004ȑǹTay ک CaoȐ2001ȑࣴزࡰрǴӭ໨Ԅ ਡЈӧ૽ግ໘ࢤሡाၨߏޑਔ໔Ǵځ่݀όᓬܭ RBFǴࡺҁࣴزᒧ᏷ RBF ࣁਡ ЈڄኧǶ

2.1.2 ᆛ਱ཛྷ൨Ȑgrid-searchȑϷҬΰᡍ᛾ȐCross Validationȑ

ӧRBF ਡЈڄኧύԖٿঁୖኧǴϩձᚵᆦॶȐpenalty value, C炸 Ϸୖኧ

γ

^Ǵ

Ѹ໪ᒧ᏷΋ಔӳޑୖኧٰှ،ୢᚒǶҁࣴز٬Ҕᆛ਱ཛྷ൨Ȑgrid-searchȑǴ೸ၸׯ ᡂ໔ຯෳ၂؂ჹȐC,

γ

ȑϐᆒዴࡋǴࣁቚуᆛ਱ཛྷ൨ਏ౗Ǵҁࣴزаࡰኧࠠ

Ȑexponentialȑቚуޑׇӈ଺ෳ၂ǴӵǺC = 2^í5, 2^í3, …, 2¹⁵ǹ

γ

^{= 2í15, 2í13, …, 23Ƕ}

ҁࣴز٠೸ၸҬΰᡍ᛾ᑔᒧрനଯᆒྗࡋਔϐୖኧჹȐC,

γ

^{ȑǴᆀࣁന٫ୖኧ}

ჹȐbest pair of parametersȑǴځύҔܭ૽ግၗ਑ࡌҥኳࠠޑኬҁη໣ᆀࣁ૽ግ໣

ȐTraining SetȑǴԶഭΠҔаᡍ᛾ኳࠠϐኬҁη໣߾ᆀࣁᡍ᛾໣ȐValidation SetȑǶ ߥ࡭ҬΰᔠۓݤȐHoldout Cross ValidationȑǴջஒচۈၗ਑໣ᒿᐒჄϩԋٿ

ঁᐱҥޑη໣ӝǴҭࣁ૽ግ໣ᆶᡍ᛾໣Ǵ٠ѝ٬Ҕ૽ግ໣ࡌҥኳࠠǴӆ٬Ҕᡍ᛾

໣Ⴃෳᒡрॶ٠ղᘐځ҅ዴ܄Ƕ

K ԛҬΰᡍ᛾ȐK-fold Cross Validationȑޑբݤࢂஒ૽ግ໣֡ϩԋ K ҽǴ؂

ԛڗځύ΋ҽ଺ࣁᡍ᛾໣ǴഭΠK-1 ҽ଺ࣁ૽ግ໣Ǵख़ፄ؁ᡯᡍ᛾ K ԛǶ

2.2 షౄંତȐconfusion matrixȑ

೸ၸషౄંତȐconfusion matrixȑஒᄽᆉݤޑ܄ૈຎ᝺ϯǴࢂ΋ᅿ಍ीኳࠠ

ຑ՗ޑ኱ྗπڀǶځύ؂ՉȐcolumnȑж߄΋ঁჴሞϩᜪǹԶӈȐrowȑ߾߄Ң ႣෳϩᜪǴ่݀཮೸ၸႣෳॶᆶჴሞॶϐᜢ߯Ǵஒኳࠠύޑ܌Ԗਢٯϩᜪډόӕ ޑᜪձҞᒵǶӵ߄ 2-1 ܌ҢǴ٠೸ၸྗዴࡋȐAccuracyȑǵᆒዴࡋȐPrecisionȑǵ єӣ౗ȐRecallȑϷ F1-measure ٰຑ՗ෳໆȐEvaluation MeasurementȑȐKohavi &

Provost, 1992; Powers et al., 2011ȑǶ

ᏃᆅྗዴࡋӕਔԵໆ੿໚܄ȐTrue positiveȑǵ੿഍܄ȐTrue negativeȑႣෳϩ ᜪ่݀ǴܭᕉნԦࢉሦୱᔈҔਔǴаᆒዴࡋǵєӣ౗ǵF1-measure ฻ࡰ኱ीᆉ߄

ၲǴё٬Ԧࢉϩᜪ่݀׳మධȐӵǺPandey et al., 2013ȑǶ

߄ 2-1 షౄંତȐConfusion matrixȑ ჴሞϩᜪ

Ⴃ

ෳ ϩ ᜪ

໚܄

ȐCondition Positiveȑ

഍܄

ȐCondition Negativeȑ Test Positive ੿໚܄

ȐTrue positiveȑ

ଵ໚܄

ȐFalse positiveȑ Test Negative ଵ഍܄

ȐFalse negativeȑ

੿഍܄

ȐTrue negativeȑ ᕴໆȐTotal populationȑ!

¾ ྗዴࡋȐ……—”ƒ…›ȑ ൌσ ୘୰୳ୣ୮୭ୱ୧୲୧୴ୣାσ ୘୰୳ୣ୬ୣ୥ୟ୲୧୴ୣ

σ ୘୭୲ୟ୪୮୭୮୳୪ୟ୲୧୭୬

¾ ᆒዴࡋȐ”‡…‹•‹‘ȑ ൌσ ୘୰୳ୣ୮୭ୱ୧୲୧୴ୣ

σ ୘ୣୱ୲୔୭ୱ୧୲୧୴ୣ

¾ єӣ౗Ȑ‡…ƒŽŽȑ ൌ σ ୘୰୳ୣ୮୭ୱ୧୲୧୴ୣ

σ େ୭୬ୢ୧୲୧୭୬୔୭ୱ୧୲୧୴ୣ

¾ ͳ െ ‡ƒ•—”‡ ൌଶൈୖୣୡୟ୪୪ൈ୔୰ୣୡ୧ୱ୧୭୬

ୖୣୡୟ୪୪ା୔୰ୣୡ୧ୱ୧୭୬

2.3 ϣఘᛥࡰ኱ȐNemerow Index, P

N

ȑᆶϩભ

ϣఘᛥࡰ኱ȐNemerow Index, PNȑҔܭຑሽᕉნࠔ፦ਔǴό໻Եቾຑ՗ϐᅱ

ෳ໨ѳ֡ॶǴҭуख़ीᆉຬ኱ำࡋനε໨Ǵёँᡉຬ኱ำࡋനεϐ໨ҞቹៜǶ ȐNemerow, 1974ǹ஭Ǵ2012ȑǶ

Ҟ߻ᆵ᡼ᕉߥ࿿ςࡌ࿼ӭ໨ᕉნࡰ኱ᅱෳس಍ǴٯӵǺޜ਻Ԧࢉᆕӝࡰ኱

ȐPSIȑǵН፦ࡰኧȐWQIȑϷݞοԦࢉࡰኧȐRPIȑ฻Ƕՠ RPI ղۓݞοН፦Ԧ

ࢉำࡋϐղᘐ୷ྗ໻ࣁѤϩݤǴ٠όԵໆচҁόӕНᡏϩᜪϐཀ఼ǶશȐ2010ȑ

ࡰрНྍߥៈ୔ϐݞࢤᅱෳǴϿኧН፦ୖኧຬၸ኱ྗॶਔǴӧН፦ࡰ኱΢٠คᡉ

๱ϸᔈǴჴѨѐຑ՗٠௦ڗ፾྽ᆅڋ௛ࡼϐӃᐒǹฅԶǴPNёаղᘐᕉნ፦ໆᆶ ຑሽ኱ྗϐ໔ޑᜢ߯Ǵ೸ၸᒧ᏷፾྽ޑຑ݋୷ྗॶȐCsiȑፓ፾ϐǶԖᜢβᝆԦࢉ

Ў᝘ύࡰрǴऩᒧҔ၀୯βᝆङඳॶࣁCsiǴຑሽࡕϐPNኧॶ೯தλܭ3ǹԶ PN

εܭ3 ޑӦ୔Ǵ൳ЯዴۓࣁԦࢉ୔ୱȐቅ฻Ǵ2009ǹ஭฻Ǵ2012ȑǶ

PNឦܭᆕӝࡰ኱Ǵၨൂ΋ࡰ኱ܰΑှ᏾ᡏβᝆख़ߎឦԦࢉǶҞ߻ᆵ᡼ᕉߥ࿿

ुۓӚ໨ख़ߎឦȐઈǵᙿǵሐǵልǵ؄ǵᙻǵႉǵᎋȑϐβᝆᆅڋ኱ྗϷᅱෳ኱

ྗǴаԜڰۓߐᘖբࣁຑ՗୷ྗឦܭൂ΋ࡰ኱БݤǴၨᜤޕځԦࢉӄᇮǶ೚ӭࣴ

زගډPN཮ӢӦ୔ǵङඳόӕϷຑሽჹຝޑׯᡂԶԖৡ౦Ǵӵຑ՗ഌୱНᡏϷβ ᝆԦࢉǴ߾ԖόӕޑۓကǴሡ٩ྣࣴزჹຝගр፾྽ޑϩᜪǴ٠ଞჹৡ౦฻చҹ ᒧۓ฻ભϩᜪ኱ྗȐӵࢫ฻Ǵ2012ǹࢫ฻Ǵ2013ǹቅ฻Ǵ2014ȑǶ

аΠ܌ӈࣁӚۓӦ೸ၸϣఘᛥᆕӝࡰ኱ຑሽځᕉნ፦ໆڋۓϐϩᜪ኱ྗǴӵ ߄ 2-2ǴՏܭύ୯໦ࠄ࣪೯ੇᑜޑ׻᜾෫ຑሽӦ߄НǴӢځӦ߄Нޑ፦ໆςࣁύ ୯ᡉ๱ᕉნೕჄکᆅ౛ޑୢᚒϐ΋ȐTang et al., 2011ȑǹϷ߄ 2-3Ǵࣁӧύ୯ϼՉ ξܿ᜾܌ᅱෳӦΠНޑϩᜪ኱ྗȐ׵฻Ǵ2009ȑǹќѦǴ୯ϣѦᔈҔӧβᝆၭӦख़ ߎឦϐϩભǴӵ߄ 2-4 ܌ҢȐ஭฻Ǵ2012ȑǶ

߄ 2-2 PNӦ߄Нϐϩભ኱ྗ

PNϩભ ʀ 1 1 ~ 2 2 ~ 3 3 ~ 5 ʁ 5

Ӧ߄Н฻䵠 మዅ ᇸࡋԦࢉ Ԧࢉ ख़ࡋԦࢉ ᝄख़Ԧࢉ

ȐTang et al., 2011ȑ

߄ 2-3 PNӦΠНϐϩભ኱ྗ

PNϩભ ʀ 0.8 0.8 ~ 2.5 2.5 ~ 4.25 4.25 ~ 7.2 ʁ 7.2

ӦΠН฻䵠 ᓬؼ ؼӳ ၨӳ ၨৡ ཱུৡ

Ȑ׵฻Ǵ2009ȑ

߄ 2-4 PNβᝆၭӦख़ߎឦϩભ኱ྗ

PNϩભ < 0.7 0.7 ~ 1.0 1.0 ~ 2.0 2.0 ~ 3.0 ʁ 3.0

βᝆԦࢉ฻䵠 ᓬؼ Ӽӄ ᝾י Ԧࢉ Ӓ্

Ȑ஭฻Ǵ2012ȑ

ಃΟകġ ࣴز׷਑ǵБݤϷࢬำ

3.1 ࣴز׷਑

3.1.1 ࣴز୔ୱ

ᄒԿ 102 ԃ҃Ǵ٩ᕉߥ࿿Ϧ֋ၗ਑ᡉҢǴಕीӚᑜѱၭӦ೏ፓࢗЪϦ֋ӈᆅ ϐၭӦӅ 4,402 ฽Ȑय़ᑈ 746 ϦഘȑǴځύаਲ༜ѱϦ֋ϐ 1,701 ฽Ȑय़ᑈ 211 ϦഘȑࣁനӭǴځԛࣁᄆϯᑜ 328 ฽Ȑय़ᑈ 61 ϦഘȑǴࡺҁࣴزᒧ᏷ᄆϯᑜǵ ਲ༜ѱբࣁࣴز୔ୱǶ

߄ 3-1 ࣴز୔ཷॊ

ᄆϯᑜ ਲ༜ѱ

βӦय़ᑈȐϦഘȑ 107,440 122,100

હӦय़ᑈȐϦഘȑ 69,111 37,544

Ьाౢ཰ ࢉ᏾ǵ੬ށǵႝᗓǵ

ߎឦ߄य़ೀ౛฻

ߎឦᇙࠔᇙ೷཰ǵ ᐒఓ೛ഢᇙ೷অଛ

཰ǴϷႝηᐒఓᏔ

׷ᇙ೷ǵঅଛ཰฻

ࣁЬ

ၗ਑᏾౛Ծᕉߥ࿿ीฝǴ2011! ! ! ! !

Ȑ΋ȑᄆϯᑜ

ᄆϯᑜՏܭᆵ᡼࣪ύ೽Ǵӵკ 3-1 ᄆϯᑜՉࡹ୔ୱ܌ҢǴӅϩ 26 ໂᙼѱǴ чଆεغྛᆶѠύᑜࣁࣚǴࠄԿᐜНྛᆶ໦݅ᑜࢦᎃǴܿॸΖړξᆶࠄ׫ᑜ࣬႖Ǵ Ջᖏᆵ᡼ੇ৙ǶβӦय़ᑈ 107,440 ϦഘǴહӦय़ᑈ 69,111 ϦഘȐ՞ӄᑜय़ᑈ 65.5

%ȑǴࣁᆵ᡼ख़ाዿԯౢ୔ǴΨࢂၭށख़ᗺᑜѱǶฅԶǴᄆϯᑜϣλࠠπቷ႟ࢃණ թܭҖ໔ϐπၭషӝࠠᄊǴπቷฦ૶ӭаߎឦᇙࠔǵતᙃ཰Ϸ༟ጤᇙࠔ཰Ǵ୔ϣ ወӧԦࢉٰྍԖࢉ᏾ǵ੬ށǵႝᗓǵߎឦ߄य़ೀ౛฻ǶќѦǴᄆϯНճ཮ᄆϯπ բઠᗄ୔ϣϐឲ෸෤ၰܿՋΒǵΟӥ༸ǵЍጕڙख़ߎឦԦࢉᝄख़Ǵςԋࣁख़ᗺख़ ߎឦׯ๓ൺػӦ୔Ȑ஭Ǵ2010ȑǶ

კ 3-1 ᄆϯᑜՉࡹ୔ୱ

ȐΒȑਲ༜ѱ

ਲ༜ѱՏܭᆵ᡼࣪ՋчǴӵკ 3-2 ਲ༜ѱՉࡹ୔ୱ܌ҢǴӄѱӅჄϩࣁ 13 ୔Ǵ Ջᖏᆵ᡼ੇ৙ǴܿࠄаၲᢀξǵᙪᙪξᆶѠчϷەើጕϩࣚǴࣁᆵчࣧӦᆶਲ༜

ѠӦޑϺฅჄ୔ǶӄᑜβӦᕴय़ᑈऊ 122,100 ϦഘǴહӦय़ᑈ 37,544 ϦഘȐ՞

ӄᑜय़ᑈ 30.7 %ȑǶѱϣπቷаߎឦᇙࠔᇙ೷཰ǵᐒఓ೛ഢᇙ೷অଛ཰ǴϷႝη ᐒఓᏔ׷ᇙ೷ǵঅଛ཰฻ࣁЬǶځห௦ࣳҡ٠ӣ༤Ԗ্٣཰ቲకނǴᏤठ၀୔ୱ ϐβᝆԦࢉ׳ࣁፄᚇȐ஭Ǵ2011ǹᕉߥ࿿Ǵ2011ǹਲ༜ᑜࡹ۬Ǵ1998ȑǶ

კ 3-2 ਲ༜ѱՉࡹ୔ୱ

3.1.2 ၭӦፓࢗၗ਑

Ծ1992 ԃଆǴՉࡹଣၭ཰ہ঩཮ၭ཰၂ᡍ܌Ȑᙁᆀၭ၂܌ȑǴࣁၭӦࠔ፦౜

ݩ຾Չڀᡏፓࢗ٠ϩ݋Ҟ߻હբБԄǴ٬୯ΓΑှၭӦβᝆޥΚǵβᝆԦࢉ฻ᅪ ቾǴ٠යࡑૈٛጄԦࢉ୔ୱǵׯ๓ၭҖӦΚࠔ፦Ǵࡺځ຾ՉӄᆵၭӦय़ᑈऊ 78

࿤Ϧഘϐβᝆࠔ፦ϷғౢΚፓࢗǶፓࢗБԄа໔႖250 ϦЁࣁൂՏϐᆛ਱Ԅ௦໣

βᝆኬҁǴа0.1 M HCl ๧ڗБݤᔠෳᙿȐCdȑǵሐȐCrȑǵልȐCuȑǵᙻȐNiȑǵ

ႉȐPbȑǵᎋȐZnȑǴӅϤ໨ख़ߎឦᐚࡋǶࣁࣴ݋ၭӦख़ߎឦଯԦࢉወ༈୔ୱǴҁ

ࣴز௦Ҕᄆϯᑜ 10,371 ฽ǵਲ༜ѱ 9,005 ฽ϐ߄βኧᏵǴ߄ 3-2ǵ߄ 3-3 ϩձ ࣁᄆϯᑜǵਲ༜ѱၭӦ߄ቫβᝆޑख़ߎឦፓࢗᐚࡋ୷ҁ಍ीϩ݋Ƕ

߄ 3-2 ᄆϯᑜၭӦ߄βख़ߎឦᐚࡋፓࢗ಍ी

໨Ҟ Cd Cr Cu Ni Pb Zn

ѳ֡ॶ 0.28 0.71 10.88 3.56 8.07 37.54

ύՏኧ 0.18 0.31 7.46 2.17 6.44 8.98

ಃ1 ѤϩՏኧ 0.10 0.19 5.01 1.47 4.36 5.98

ಃ3 ѤϩՏኧ 0.37 0.52 10.40 3.34 9.79 14.04 നεॶ 89.14 363.07 5512.46 911.32 2674.47 188590.86

኱ྗৡ 1.07 5.10 59.03 12.13 28.22 1871.12

ၗ਑ᕴኧ 10,371 ฽ǴൂՏǺmgڄkg^-1ǴᔠෳൂՏǺၭ၂܌

߄ 3-3 ਲ༜ѱၭӦ߄βख़ߎឦᐚࡋፓࢗ಍ी

໨Ҟ Cd Cr Cu Ni Pb Zn

ѳ֡ॶ 0.17 0.39 18.93 1.33 8.29 18.60 ύՏኧ 0.13 0.24 4.38 0.77 5.89 7.78

ಃ1 ѤϩՏኧ 0.07 0.14 2.65 0.48 4.12 3.93

ಃ3 ѤϩՏኧ 0.21 0.40 8.57 1.30 8.47 15.94 നεॶ 24.37 40.49 6411.70 158.09 2200.25 3206.07

኱ྗৡ 0.45 0.90 129.25 3.47 29.39 59.65

ၗ਑ᕴኧ 9,005 ฽ǴൂՏǺmgڄkg^-1ǴᔠෳൂՏǺၭ၂܌

3.2 ࣴزБݤ

3.2.1 ϣఘᛥࡰ኱ȐPNȑ

ҁࣴزᒧҔϣఘᛥࡰ኱ȐNemerow index, PNȑ຾Չβᝆख़ߎឦԦࢉϐຑሽǴ

ीᆉϦԄӵǺ

»¼

« º

¬

ª +

= +

=

¦

= n 1 i

2 s 2 m si 2 i

max 2 avg

N )

C (C C )

(C n 1 2 2 1 / ) P (P

P

ԄύǴ

¦

=

= ⁿ

1

i si

i

avg )

C (C n

P 1 ǹ

¸¸¹

¨¨ ·

©

= §

si i s2

2 s1

1

max C

,C ...

C , ,C C max C

P = )

C (C

s

m ǹ

CiǺiᅱෳ໨Ҟޑჴෳॶǹ CsiǺiᅱෳ໨Ҟޑຑሽ୷ྗॶǹ nǺڙຑ՗ᅱෳ໨Ҟޑᕴኧǹ

ၭ၂܌а0.1 M HCl๧ڗβᝆख़ߎឦၗ਑ǴᆶȨᆵ᡼Ӧ୔βᝆख़ߎឦ֖ໆ኱

ྗᆶ฻ભȩϐᔠෳБݤ࣬ӕǴࡺаȨᆵ᡼Ӧ୔βᝆख़ߎឦ֖ໆ኱ྗᆶ฻ભȩϐಃ

Οભङඳॶ΢ज़Ǵ଺ࣁᅱෳ໨Ҟޑຑሽ୷ྗॶȐCsiȑ຾ՉຑሽǴӵ߄ 3-4܌ӈǶ

߄ 3-4 PNޑຑ݋୷ྗॶȐCsiȑ

ख़ߎឦ ᙿ ሐ ል ᙻ ႉ ᎋ

CsiȐmgڄkg^-1ȑ 0.39 10.0 20.0 10.0 15.0 25.0

௦Ȩᆵ᡼Ӧ୔βᝆख़ߎឦ֖ໆ኱ྗᆶ฻ભȩϐಃΟભङඳॶ΢ज़! ! !

PN < 1.0Ǵᇥܴᕉნ፦ໆၲຑ݋኱ྗޑा؃ǹPN ʁġ1.0Ǵᇥܴᕉნ፦ໆςό

ૈᅈىຑ݋኱ྗޑा؃ǴҁࣴزۓကPN < 1.0ࣁ҂ԦࢉၭӦа഍܄Ȑnegativeȑ߄ ҢǴԶPN ʁġ1.0ࣁԦࢉၭӦа໚܄Ȑpositiveȑ߄ҢǴӵ߄ 3-5Ƕ

߄ 3-5 ၭӦख़ߎឦϩભ኱ྗ

PNϩભ < 1.0 ^ʁ 1.0

฻䵠 ҂Ԧࢉ Ԧࢉ

3.2.2 Ѝ࡭ӛໆᐒȐSVMȑ

Ѝ࡭ӛໆᐒȐSupport Vector Machine, SVMȑࣁ΋ᅿᅱ࿎ԄᏢಞȐSupervised LearningȑǶ྽Ԗၗ਑ाϩԋٿᜪ܈׳ӭᜪձǴӵკ 3-3 Ѝ࡭ӛໆᐒཷۺҢཀǴᙖ җSVM Бݤפрຬѳय़ȐHyperplaneȑஒٿᅿόӕᜪձޑၗ਑ϩဂǴќѦஒຬѳ य़ᆶനௗ߈ޑ૽ግၗ਑ᗺ໔ޑຯᚆۓကࣁ໔႖ȐMarginȑǶЍ࡭ӛໆᐒޑҞ኱ջ൨ פ ڀ Ԗ ന ε ໔ ႖ ޑ ຬ ѳ य़ բ ࣁ ϩ ᜪ ᜐ ࣚ Ǵ Ԝ ਔ ᆀ ࣁ ന ٫ ຬ ѳ य़ ȐOptimal HyperplaneȑǴૈஒ૽ግၗ਑ᅰໆܴዴӦ୔႖໒ٰǶޜ໔ύന᎞߈ന٫ຬѳय़ޑၗ

਑ᗺᆀࣁЍ࡭ӛໆȐSupport VectorȑǴёճҔԜຬѳय़ղۓཥ຾ၗ਑ϐ܌ឦᜪձǶ ȐBoser et al., 1992; Cortes & Vapnik, 1995ȑ

კ 3-3 Ѝ࡭ӛໆᐒཷۺҢཀ

೛૽ግၗ਑߄ҢࣁD= i

{ }

ⁿ

n n i i i i i

i y x x x R y

x , } | ( ,..., ) | 1, 1

{ ⁿ₌₁ˢ∀ = ^{(1) ( )} ∈ ∈ + − Ǵຬѳय़

ޑኧᏢ׎Ԅёа߄ҢࣁǺ

=0

−

⋅x b w

ځύx ࢂຬѳय़΢ޑᗺǹw ࢂࠟޔܭຬѳय़ޑӛໆǹb ࣁ΋தኧǶ

ӵკ 3-4 നε໔႖ຬѳय़ǴѳՉޑ٠ЪᚆЍ࡭ӛໆന߈ޑຬѳय़җаΠБำ

௼߄ҢǺ

ቊw⋅x−b= ൅ ͳ

=

−

⋅x b

w െ ͳ

კ 3-4 നε໔႖ຬѳय़

ٿঁຬѳय़ϐ໔ޑຯᚆࢂ2/|w|ǴҞ኱؃നλϯ |w|Ƕ೬܄໔႖ȐSoft marginȑ ύǴ೸ၸуΕ᚞ԅୖኧ ξ_iȐSlack VariableȑаᑽໆჹኧᏵ x_iޑᇤϩᜪࡋǴёа

ೀ౛኱૶ᒱᇤޑ૽ግၗ਑Ǵ୔ϩ҅ॄٯޑຬѳय़όӸӧਔǴ߾ஒᒧ᏷΋ঁຬѳय़ Ꮓёૈమධޑ୔ϩၗ਑Ǵӕਔ٬ځᆶϩࣚനమධޑኬҁޑຯᚆനεϯǴЪӕਔჹ ܭ܌Ԗޑi ᅈىǴҞ኱ڄኧёа߄ҢࣁǺ

i w x b i

y

( ⋅ + )≥1−ξ 1≤i≤n

ᒿࡕǴӧቚε໔ຯکᕭλᚵᆦॶȐpenalty value, C炸ٿεҞ኱ϐ໔຾Չ៾ᑽ ᓬϯǴځύC>0Ǵӵ݀ᚵᆦॶࣁጕ܄ڄኧǴ߾߄Ңࣁ

¦

=

+ ⁿ

i i T

b w

C w

Min

w

1 ,

, 2

1 ξ

ξ

≥0 ξ

ࣁΒԛೕჄന٫ϯୢᚒȐQuadratic Programming Optimization ProblemȑǴ೯ၸ LIBSVMȐChang & Lin, 2011ȑ؃ှǴளຬѳय़ޑϩᜪղձڄኧࣁǺ

¸¸¹

¨¨ ·

©

§ »¼

« º

¬

ª +

= ¦

= N

i

i j

iy x x b

x f

1

) , ( sgn

)

( α

ჹܭߚጕ܄ୢᚒǴऩਡڄኧK(x_i,x_j)ᅈىMercerచҹK(x_i,x_j)=φ (x_i)^T(x_j)Ǵ

૽ግӛໆx_i೸ၸڄኧφࢀ৔ډଯᆢࡋޜ໔Ǵӧଯᆢࡋޜ໔פ൨ጕ܄ന٫ຬѳय़Ǵ ջளଯᆢޜ໔ޑϩᜪڄኧȐYang et al., 2008ȑࣁǺ

¸¸¹

¨¨ ·

©

§ »¼

« º

¬

ª +

= ¦

= N

i

i j

iy K x x b

x f

1

) , ( sgn

)

( α

΢ԄǴஒଯᆢޜ໔ޑϣᑈीᆉᙯඤࣁեᆢࡋڄኧीᆉǴԜϩᜪڄኧᏢಞᐒᆀ ࣁЍ࡭ӛໆᐒǴځύK(x_i,x_j)≡φ(x_i)^Tφ(x_j)ࣁਡЈڄኧȐkernel fuctionȑǴҁࣴز ᒧ᏷RBFࣁਡЈڄኧǴ߄ҢࣁǺ

) exp(

) ,

(x_i x_j x_i x_j ²

K = −γ − γ ∈ℜ⁺

3.3 ࣴزࢬำ

3.3.1 Ѝ࡭ӛໆᐒȐSVMȑϩᜪ

Ѝ࡭ӛໆᐒஒς኱ᠸȐlabelȑԋ໚܄Ϸ഍܄ޑ૽ግၗ਑໣Ȑtraining data setȑ

଺૽ግࡌኳǴځࡕǴаᡍ᛾໣ၗ਑ȐValidation data setȑ܈ཥၗ਑໣ȐNew data setȑ ٩၀ኳࠠ೸ၸSVMႣෳϩᜪ่݀٠ϩ݋ᡍ᛾Ǵࢬำӵკ 3-5Ƕ

კ 3-5Ѝ࡭ӛໆᐒȐSVMȑϩᜪࢬำ

3.3.2 ᆛ਱ཛྷ൨

җკ 3-5 ܌ҢǴа SVM ૽ግǵࡌҥኳࠠ໪๏ϒୖኧჹȐC,

γ

^{ȑǴࣁᑔᒧр}

ന٫ୖኧჹȐbest pair of parametersȑǴҁࣴز೸ၸ 10-Ҭΰᡍ᛾Ȑ10-fold cross validationȑ຾Չᆛ਱ཛྷ൨Ȑgrid-searchȑ຾Չෳ၂ୖኧჹᑔᒧǴᏹբࢬำӵკ 3-6Ƕ

კ 3-6ᆛ਱ཛྷ൨ࢬำ

ಃѤകġ ่݀ᆶ૸ፕ

4.1 ϣఘᛥࡰ኱ȐP

N

ȑຑ݋่݀

ҁࣴز٬Ҕၭ၂܌ᔠෳϐၭӦख़ߎឦၗ਑௦ኬਔ໔ࣁ҇୯81ԃԿ97ԃǴа ᄆϯᑜǵਲ༜ѱࣁࣴز୔ୱǴ೸ၸϣఘᛥࡰ኱ȐPNȑᆕӝीᆉϤ໨ख़ߎឦȐCdǵ CrǵCuǵNiǵPbǵZnȑᐚࡋၗ਑Ǵ่ӝፓࢗᗺՏޑޜ໔০኱Ȑ_௜ǡ _௜ȑǴᙖշӦ౛

ၗૻس಍Ȑgeographic information system, GISȑ৖Ңޜ໔ϩѲ٠٩஭฻Ȑ2012ȑ ࡌ᝼ϐϖભϩᜪև౜Ǵ܌ᛤࣴز୔ୱϐPNॶޜ໔ϩѲǴӵკ 4-1ǵკ 4-2Ƕ

Ȑ΋ȑᄆϯᑜ

ᄆϯᑜຑ݋่݀ӵ߄ 4-1܌ҢǴ 25.22 %Ȑ6,461฽ȑឦܭ ȨᓬؼȐPN ɦ 0.7ȑȩǹ 21.25 %Ȑ1,459฽ȑឦܭȨӼӄȐ0.7ʀġPN ɦ1.0ȑȩǹ 23.82 %Ȑ1,954฽ȑឦܭ Ȩ᝾יȐ1.0ʀġPN ɦ2.0ȑȩǹ 23.82 %Ȑ248฽ȑឦܭ ȨԦࢉȐ2.0ʀġPN ɦ3.0ȑȩǹ നࡕ 29.71 %Ȑ249 ฽ȑឦܭ ȨӒ্ȐPN ʁġ3.0ȑȩǶځύǴPN ʁġ1.0 ՞၀୔

฽ኧ 23.63 %Ǵ߄ҢβᝆԦࢉނ፦ຬၸङඳॶ΢ज़Ǵբނғౢёૈڙډख़ߎ

ឦԦࢉǶ

ȐΒȑਲ༜ѱ

ਲ༜ѱຑ݋่݀ӵ߄ 4-2܌ҢǴ 72.40 %Ȑ6,520฽ȑឦܭ ȨᓬؼȐPN ɦ 0.7ȑȩǹ 9.34 %Ȑ841฽ȑឦܭȨӼӄȐ0.7ʀġPN ɦ1.0ȑȩǹ 9.46 %Ȑ852฽ȑឦܭȨ᝾ יȐ1.0ʀġPN ɦ2.0ȑȩǹ 2.89 %Ȑ260 ฽ȑឦܭȨԦࢉȐ2.0ʀġPN ɦ3.0ȑȩǹന ࡕ 5.91 %Ȑ532฽ȑឦܭȨӒ্ȐPN ʁġ3.0ȑȩǶځύβᝆԦࢉނ፦ຬၸङඳ ॶ΢ज़ȐPN ʁġ1.0ȑ՞၀୔ 18.26 %Ƕ

߄ 4-1 ᄆϯᑜ PNϩભϩѲ߄

Ԧࢉ฻ભ PNϩભ ฽ኧ ԭϩК

ᓬؼ PNɦ0.7 6,461 62.30 Ӽӄ 0.7ʀPNɦ1.0 1,459 14.07

᝾י 1.0ʀPNɦ2.0 1,954 18.84 Ԧࢉ 2.0ʀPNɦ3.0 248 2.39

Ӓ্ PNʁ3.0 249 2.40

ᕴኧ 10,371 100

კ 4-1 ᄆϯᑜӚ௦ኬᗺ PNϩѲკ

߄ 4-2 ਲ༜ѱ PNϩભϩѲ߄

Ԧࢉ฻ભ PNϩભ ฽ኧ ԭϩК

ᓬؼ PNɦ0.7 6,520 72.40 Ӽӄ 0.7ʀPNɦ1.0 841 9.34

᝾י 1.0ʀPNɦ2.0 852 9.46 Ԧࢉ 2.0ʀPNɦ3.0 260 2.89 Ӓ্ PNʁ3.0 532 5.91

ᕴኧ 9,005 100

კ 4-2 ਲ༜ѱӚ௦ኬᗺ PNϩѲკ

4.1.1 ૽ግ໣ኬҁၗ਑኱ᠸ Ȑ΋ȑᄆϯᑜ

Ѝ࡭ӛໆᐒȐSVMȑሡஒ૽ግၗ਑኱ᠸࡕ຾ՉϩᜪǴӧԜஒᄆϯᑜ܌Ԗ௦ኬ ᗺPNǴа1.0 բࣁሚॶȐthreshold valueȑ୔ϩࣁٿᜪǶ೸ၸ GIS ᛤрǴӵკ 4-3Ƕ ځύǴ 2,451 ฽ឦܭPN ʁ 1.0Ȑ՞ 23.63 %ȑǴ኱ᠸࣁ໚܄ȐpositiveȑǹаϷ 7,920

฽ឦܭPN ɦ 1.0Ȑ՞ 76.37 %ȑǴ኱ᠸࣁ഍܄ȐnegativeȑǴӵ߄ 4-3Ƕ

߄ 4-3 ᄆϯᑜ PNϩભȐሚॶ೛ۓࣁ1.0ȑ PNϩભ ኱ᠸ ฽ኧ ԭϩКȐ%ȑ PNʁ1.0 ໚܄ 2,451 23.63 PNɦ1.0 ഍܄ 7,920 76.37

ᕴኧ 10,371 100

კ 4-3 ᄆϯᑜሚॶࣁ 1.0 ϐӚ௦ኬᗺ PN

ȐΒȑਲ༜ѱ

२ӃǴӕኬӦஒਲ༜ѱ܌Ԗ௦ኬᗺаPN ࣁ 1.0 բࣁሚॶȐthreshold valueȑ

୔ϩࣁٿᜪǴӵკ 4-4ǶځύǴ 18.26 %Ȑ1,644 ฽ȑឦܭPN ʁ 1.0ǹ 81.74 % Ȑ7,361 ฽ȑឦܭPN ɦ 1.0Ǵӵ߄ 4-4Ƕ

߄ 4-4 ਲ༜ѱ PNϩભȐሚॶ೛ۓࣁ1.0ȑ PNϩભ ኱ᠸ ฽ኧ ԭϩКȐ%ȑ PNʁ1.0 ໚܄ 1,644 18.26 PNɦ1.0 ഍܄ 7,361 81.74

ᕴኧ 9,005 100

კ 4-4 ਲ༜ѱሚॶࣁ 1.0 ϐӚ௦ኬᗺ PN

4.2 ૽ግ໣኱ᠸಔԋКٯ௖૸

ҁ࿯ڰۓኬҁύ኱ᠸࣁ໚܄Ȑpositiveȑޑၗ਑฽ኧǴׯᡂ኱ᠸࣁ഍܄Ȑnegativeȑ ޑၗ਑ᒧڗໆǴаόӕКٯᒿᐒܜڗኧໆಔԋ૽ግ໣Ȑtraining data setȑǴ٠а၀

୔ୱᕴၗ਑ᗺෳ၂ኳࠠȐmodelȑਏૈǴ௖૸૽ግ໣ኬҁಔԋКٯჹෳ၂่݀ޑቹ ៜǶᄆϯᑜǵਲ༜ѱޑਡЈڄኧୖኧჹȐC,

γ

^{ȑϩձ೛ࣁۓॶȐ0.5, 256.0ȑϷȐ0.125,}

256.0ȑǶෳ၂่݀ϩձӵ߄ 4-5ǵ߄ 4-6 ܌ҢǶᔠෳ่݀ᡉҢǴӧ྽૽ግ໣ኬҁ

኱ᠸ໚܄ǵ഍܄КǴၨ౛གྷϐ่݀วғܭǴᄆϯᑜ໚܄ǵ഍܄Кࣁ1Ǻ2ਔF1-measure ࣁ 0.692 ǹ ਲ༜ѱ໚܄ǵ഍܄Кࣁ1Ǻ1ਔǴ F1-measureࣁ 0.497Ƕ

߄ 4-5ᄆϯᑜኬҁКٯჴᡍ่݀

1Ǻ1 1Ǻ2 1Ǻ3

PǺPositive 2451 2451 2451

NǺNegative 2451 4902 7353

True positive 1924 1706 1444

False negative 527 745 1007

True negative 6524 7148 7421

False positive 1396 772 499

ྗዴࡋȐAccuracyȑ! 81.46% 85.37% 85.48%ġ єӣ౗ȐRecallȑ! 78.50% 69.60% 58.91%ġ ᆒዴࡋȐPrecisionȑ! 57.95% 68.85% 74.32%ġ

F1-measure 0.667 0.692 0.657ġ

߄ύԪۭࣁF1-measure നଯ໨! ! ! ! ! ! !

߄ 4-6ਲ༜ѱኬҁКٯჴᡍ่݀

1Ǻ1 1Ǻ2 1Ǻ3 1Ǻ4

PǺPositive 1644 1644 1644 1644

NǺNegative 1644 3288 4932 6576

True positive 1310 486 292 0

False negative 334 1158 1352 1644

True negative 5136 6961 7172 7361

False positive 2225 400 189 0

ྗዴࡋȐAccuracyȑ 71.58% 82.70% 82.89% 81.74%

єӣ౗ȐRecallȑ 79.68% 29.56% 17.76% 0.00%

Precision 37.06% 54.85% 60.71% 0.00%

4.3 ૽ግ໣ኬҁኧໆ่݀௖૸

ҁ࿯௖૸૽ግ໣ኬҁኧໆᡂϯჹܭ่݀ޑቹៜǶਥᏵ 4.2 ่݀Ǵ૽ግ໣ኬҁ ኧໆ໚܄ǵ഍܄Кٯܭᄆϯᑜڰۓࣁ1Ǻ2ǴԶਲ༜ѱڰۓࣁ1Ǻ1Ƕ

аӚ௦ኬᗺ০኱Ȑܺ_௜ǡ _௜ȑբࣁ૽ግϐ઩ЇᜪձǴᒧۓRBFࣁਡЈڄኧǴ٩ׇ

ڗόӕКٯᒿᐒܜڗኧໆޑ૽ግၗ਑໣ኬҁȐ10 %, 20 %, …, 100 % ૽ግ໣ȑǴ೸

ၸᆛ਱ཛྷ൨Ȑgrid-searchȑམଛ10-Ҭΰᡍ᛾Ȑ10-fold cross validationȑǴᑔᒧന٫

ୖኧჹȐbest pair of parametersȑǴаᕴၗ਑໣ȐTotal data setȑႣෳǴ௖૸ځྗዴ ࡋȐAccuracyȑǵєӣ౗ȐRecallȑǵᆒዴࡋȐPrecisionȑǵF1-measure ϷԦࢉወ༈

ጄൎ฻่݀Ƕ

ᄆϯᑜǵਲ༜ѱ૽ግ໣ኧໆᆶႣෳ่݀ϩձӵ߄ 4-7 Ϸ߄ 4-8ǹ૽ግ໣ኧໆ ᆶᆛ਱ཛྷ൨ന٫ୖኧჹပӧკύᆘՅ୔໔Ǵᡍ᛾ϷႣෳ่݀ӵ߄ϩձӵკ 4-5ǵ კ 4-7ǹ૽ግ໣ኧໆᆶଯԦࢉወ༈კϩձӵკ 4-6ǵკ 4-8 ܌ҢǶ่݀ᡉҢǴ྽

૽ግ໣ኬҁኧໆຫଯǴኳࠠႣෳ่݀F1-measureຫଯǴж߄SVMޑϩᜪਏૈᒿ ᏢಞኬҁኧቚӭԶගଯǶ

ᄆϯᑜа7,353฽ᗺՏӧ໚܄ǵ഍܄኱ᠸК1Ǻ2Π૽ግࡌҥϐኳࠠ຾Չβᝆ

ख़ߎឦԦࢉወ༈ႣෳǴ٠аᕴၗ਑ෳ၂่݀ᄆϯᑜଯԦࢉወ༈ᗺՏኧࣁ 2,478 ฽ Ȑ՞ 23.89 %ȑǴྗዴࡋȐAccuracyȑࣁ85.37%ǵF1-measureࣁ0.692ǴЬाᆫ໣

ϩѲܭчᄆϯکऍໂǵࠄᙑᐜНྛࢬୱ΢ෞǶ

ਲ༜ѱӧ኱ᠸКࣁ1Ǻ1ΠǴа3,288฽ၗ਑૽ግࡌҥኳࠠǴ٠аᕴၗ਑ෳ၂

่݀ଯԦࢉወ༈ᗺՏኧࣁ 3,535 ฽Ȑ՞ 39.25 %ȑǴྗዴࡋࣁ 71.58 %ǵF1-measure

ࣁ0.506Ǵᆫ༧ϩѲණܭਲ༜୔ୱϣǶ

Ъӧλኬҁኧ 10 % ૽ግਔᄆϯᑜǵਲ༜ѱF1-measureϩձၲ0.611Ϸ0.421Ǵ ᆶ୔ୱӚձᕴ૽ግၗ਑ໆޑF1-measure࣬КǴ߄ҢԜࣴزਢٯޑSVMӧե૽ግ

28

߄ 4-7ᄆϯᑜόӕ૽ግ໣ኧໆჹᔈႣෳ่݀ ኬҁኧໆ 10% 20% 30% 40% 50% 60% 70% 80% 90% 100% ȐPNɪ1.0ȑ2454907359801225ġ14701715196022052451ġ ȐPNɦ1.0ȑ490980147019612451ġ29413432392244124902ġ C0.5 2.0 8.0 2048.0 8192.0 0.5 8192.0 2.0 2.0 2.0

γ

16.0 16.0 16.0 16.0 16.0 256.0 16.0 256.0 256.0 256.0 1519 1441 1498 1636 1614 1585 1583 1666 1718 1706 932 1010 953 815 837 866 868 785 733 745 6921 7215 7118 6731 7098 7235 6769 7170 7086 7148 999 705 802 1189 822 685 1151 750 834 772 Accuracyȑ81.38%83.46%83.08%80.68%84.00%ġ85.04%80.53%85.20%84.89%85.37%ġ Recallȑ61.97%58.79%61.12%66.75%65.85%ġ64.67%64.59%67.97%70.09%69.60%ġ Precisionȑ60.33%67.15%65.13%57.91%66.26%ġ69.82%57.90%68.96%67.32%68.85%ġ 10.6270.6310.6200.661ġ0.6710.6110.6850.6870.692ġ ૽ግ໣Ǻ 7,353 ฽ǹႣෳ໣Ǻ 10,371 ฽

29

Ȑaȑ10 % ૽ግ໣Ȑbȑ20 % ૽ግ໣ Ȑcȑ30 % ૽ግ໣Ȑdȑ40 % ૽ግ໣ კ 4-5ᄆϯᑜόӕ૽ግ໣ኧໆᆶᆛ਱ཛྷ൨่݀

30

Ȑeȑ50 % ૽ግ໣Ȑfȑ60 % ૽ግ໣ Ȑgȑ70 % ૽ግ໣Ȑhȑ80 % ૽ግ໣ კ 4-5ᄆϯᑜόӕ૽ግ໣ኧໆᆶᆛ਱ཛྷ൨่݀Ȑុȑ

31

Ȑiȑ90 % ૽ግ໣Ȑjȑ100 % ૽ግ໣ კ 4-5ᄆϯᑜόӕ૽ግ໣ኧໆᆶᆛ਱ཛྷ൨่݀Ȑុȑ

32

Ȑaȑ10 % ૽ግ໣Ȑbȑ20 % ૽ግ໣ Ȑcȑ30 % ૽ግ໣Ȑdȑ40 % ૽ግ໣ კ 4-6ᄆϯᑜόӕ૽ግ໣ኧໆᆶଯԦࢉወ༈ϩѲࣚۓ

33

Ȑeȑ50 % ૽ግ໣Ȑfȑ60 % ૽ግ໣ Ȑgȑ70 % ૽ግ໣Ȑhȑ80 % ૽ግ໣ კ 4-6ᄆϯᑜόӕ૽ግ໣ኧໆᆶଯԦࢉወ༈ϩѲࣚۓȐុȑ

34

Ȑiȑ90 % ૽ግ໣Ȑjȑ100 % ૽ግ໣ კ 4-6ᄆϯᑜόӕ૽ግ໣ኧໆᆶଯԦࢉወ༈ϩѲࣚۓȐុȑ

35

߄ 4-8ਲ༜ѱόӕ૽ግ໣ኧໆჹᔈႣෳ่݀ ኬҁኧໆ 10% 20% 30% 40% 50% 60% 70% 80% 90% 100% PositiveȐPNɪ1.0ȑ164328493657822ġ9861150131514791644ġ NegativeȐPNɦ1.0ȑ164329493658822ġ9861151131514801644ġ C0.5 32.0 32.0 0.5 2.0 2.0 0.5 8.0 2.0 2.0

γ

16.0 16.0 1.0 256.0 16.0 256.0 16.0 256.0 256.0 256.0 positive 1191 1099 1253 1081 1185 1249 1210 1181 1212 1310 negative 453 545 391 563 459 395 434 463 432 334 negative 4535 4520 4196 5315 4762 4985 4606 4962 5226 5136 positive 2826 2841 3165 2046 2599 2376 2755 2399 2135 2225 Accuracyȑ63.59%62.40%60.51%71.03%66.04%ġ69.23%64.59%68.22%71.49%71.58%ġ Recallȑ72.45%66.85%76.22%65.75%72.08%75.97%73.60%71.84%73.72%79.68% Precisionȑ29.65%27.89%28.36%34.57%31.32%ġ34.46%30.52%32.99%36.21%37.06%ġ easure 0.4210.3940.4130.4530.437ġ0.4740.4310.4520.4860.506ġ ૽ግ໣Ǻ 3,288 ฽ǹႣෳ໣Ǻ 9,005 ฽!!!!

36

Ȑaȑ10 % ૽ግ໣Ȑbȑ20 % ૽ግ໣ Ȑcȑ30 % ૽ግ໣Ȑdȑ40 % ૽ግ໣ კ 4-7ਲ༜ѱόӕ૽ግ໣ኧໆᆶᆛ਱ཛྷ൨่݀

37

Ȑeȑ50 % ૽ግ໣Ȑfȑ60 % ૽ግ໣ Ȑgȑ70 % ૽ግ໣Ȑhȑ80 % ૽ግ໣ კ 4-7ਲ༜ѱόӕ૽ግ໣ኧໆᆶᆛ਱ཛྷ൨่݀Ȑុȑ

38

Ȑiȑ90 % ૽ግ໣Ȑjȑ100 % ૽ግ໣ კ 4-7ਲ༜ѱόӕ૽ግ໣ኧໆᆶᆛ਱ཛྷ൨่݀Ȑុȑ

39

Ȑaȑ10 % ૽ግ໣Ȑbȑ20 % ૽ግ໣ Ȑcȑ30 % ૽ግ໣Ȑdȑ40 % ૽ግ໣ კ 4-8ਲ༜ѱόӕ૽ግ໣ኧໆᆶଯԦࢉወ༈ϩѲࣚۓ

40

Ȑeȑ50 % ૽ግ໣Ȑfȑ60 % ૽ግ໣ Ȑgȑ70 % ૽ግ໣Ȑhȑ80 % ૽ግ໣ კ 4-8ਲ༜ѱόӕ૽ግ໣ኧໆᆶଯԦࢉወ༈ϩѲࣚۓȐុȑ

41

Ȑiȑ90 % ૽ግ໣Ȑjȑ100 % ૽ግ໣ კ 4-8ਲ༜ѱόӕ૽ግ໣ኧໆᆶଯԦࢉወ༈ϩѲࣚۓȐុȑ

4.4 ၭӦख़ߎឦԦࢉᜢᖄ܄ຑ݋

ҁࣴزᔈҔGISמೌᛤᇙԦࢉޜ໔ӦკǴКჹԦࢉޑޜ໔ࢬթ௃׎Ǵаޜ໔

࣬ᜢ܄ٰև౜ӭБၗ਑ޑᜢᖄ܄ϩ݋Ƕᙖ༼᏾ SVM ୔ϩၭӦख़ߎឦଯԦࢉወ༈

୔ϐޜ໔ᆶឦ܄ၗ਑Ǵ঺᠄ݞοࢬୱǵπቷǵπ཰୔ޑޜ໔ϩѲǴඓඝӚՉ཰ϐ Ԧࢉ੝ቻȐ߄ 4-9ȑǵޜ໔ϩѲǴᆕӝϩ݋ЬाԦࢉϐख़ߎឦ໨ҞǴ٠ှ݋ځԦࢉ

ԋӢǴӵკ 4-9ǵკ 4-10Ƕ

ϩ݋่݀ᡉҢǴଯԦࢉወ༈ϐ୔ୱӭኧߚଽว൑֟Ǵӭೀπቷǵπ཰୔ڬᜐ ڀଯԦࢉወ༈ǴԦࢉᒿ෤ၰࢬ୏໺ᒡᘉණޑ౜ຝǴځϐ໔ዴჴӸӧᜢᖄ܄Ǵ୤஭

Ȑ2011ȑගᒬǴа௢՗კ঺᠄πቷǴᗨё௢ෳԦࢉٰྍǴՠၭӦख़ߎឦԦࢉڀԖ ಕᑈ܄ୢᚒǴπቷ཮໒཰ଶ཰ǴཥπቷฦᐒёаᆒዴۓՏǴՠࠅߚԦࢉޑޔௗٰ

ྍǴᙑπቷёૈଶ཰ǵᙯ཰Ǵவᕉნၗ਑৤္੃ѨǴՠࠅࢂ੿҅ޑԦࢉٰྍǴа ᕉნၗ਑৤ှញୢᚒԖ΋ۓޑ॥ᓀ܄Ǵሡۓය׳ཥၗ਑৤ᆶ౜ӦፓࢗుΕΑှୢ

ᚒٰྍǶќѦǴᆶၸ۳ᜪ՟ޑࣴزКၨǴஒᑔᒧϐӒ্ϷԦࢉ฻ભޑНճλಔጄ ൎȐࢫ฻Ǵ2013ȑϷၭҖНճ཮ᑔᒧрಃIIIભӥၡڀԦࢉወ༈୔ୱȐၭہ཮Ǵ2014ȑǴ ᆶҁࣴز่݀঺᠄຾ՉКჹǶว౜ҁࣴزࡷᒧጄൎࣁ PNʁ1Ǵࡺጄൎ཮КНճλ ಔᑔᒧຑ݋ၨε΋٤Ǵගٮ׳ӭ҂ٰёૈڙቹៜޑጄൎǴӵკ 4-11ǵკ 4-12Ǵ Զаឲ෸س಍ᡉҢԦࢉወ༈୔ୱؒԖϩޑࡐಒஏǴӵԦࢉӥၡёૈࣁ᏾చӥၡ܈

ЍጕǴऩམଛҁࣴز่݀׳ૈమཱޑΑှڙԦࢉቹៜጄൎǴӵკ 4-13ǵკ 4-14Ƕ ߄ 4-9Չ཰ᜪձϷځख़ߎឦԦࢉᅿᜪ

Չ཰ᜪձ ЬाԦࢉख़ߎឦᅿᜪ

ᇙॠ཰ Cd, Cr, Cu, Pb, Zn

ߎឦ୷ҁπ཰ As, Cd, Cr, Cu, Hg, Ni, Pb, Zn ߎឦ߄य़ೀ౛཰ǵႝᗓ཰ Cd, Cr, Cu, Hg, Ni, Pb, Zn

඲༝ᇙ೷ϷъᏤᡏᇙ೷཰ As, Cd, Cr, Cu, Hg, Ni, Pb

43

კ 4-9ᄆϯᑜଯԦࢉወ༈୔ᆕӝࣴ݋კ

44

კ 4-10ਲ༜ѱଯԦࢉወ༈୔ᆕӝࣴ݋კ

45

კ 4-11ᄆϯᑜНճλಔຑ݋ԦࢉϷӒ্฻ભ঺᠄SVMޑႣෳ่݀

46

კ 4-12ਲ༜ѱНճλಔຑ݋ԦࢉϷӒ্฻ભ঺᠄SVMޑႣෳ่݀

47

კ 4-13ᄆϯᑜಃIIIભӥၡڀԦࢉወ༈୔ୱ঺᠄SVMޑႣෳ่݀

48

კ 4-14ਲ༜ѱಃIIIભӥၡڀԦࢉወ༈୔ୱ঺᠄SVMޑႣෳ่݀

ಃϖകġ ่ፕ

1. ࿶Ѝ࡭ӛໆᐒȐSupport Vector Machine, SVMȑϩᜪǴᄆϯᑜଯԦࢉወ༈ᗺՏ

ኧࣁ 2,478 ฽Ȑ՞ 23.89 %ȑǴЬाᆫ໣ϩѲܭчᄆϯکऍໂǵࠄᙑᐜНྛࢬୱ

΢ෞǹਲ༜ѱଯԦࢉወ༈ᗺՏኧࣁ 3,535 ฽Ȑ՞ 39.25 %ȑǴᆫ༧ϩѲණܭਲ༜

୔ୱϣǴԜ่݀ૈբࣁࡹ۬ࡕុԦࢉᆅ౛ᆶٛݯೕჄୖԵϐ٩ᏵǶ

2. ᄆϯᑜа7,353฽ᗺՏӧ໚܄ǵ഍܄኱ᠸК1Ǻ2Π૽ግࡌҥϐኳࠠ຾Չβᝆख़

ߎឦԦࢉወ༈ႣෳǴ่݀ྗዴࡋȐAccuracyȑࣁ 85.37%ǵF1-measure ࣁ 0.692ǹ ਲ༜ѱӧ኱ᠸКࣁ1Ǻ1ΠǴӅ3,288฽ၗ਑૽ግࡌҥኳࠠǴԦࢉወ༈Ⴃෳϐ่݀

ྗዴࡋࣁ 71.58 %ǵF1-measureࣁ0.506Ƕ

3. ҁࣴز᛾ჴаSVM ᄽᆉݤૈԖਏӦᔈҔܭβᝆख़ߎឦԦࢉወ༈ჄϩǴЪӧե

૽ግ໣ኬҁኧջёၲؼӳޑϩᜪਏૈǶ

4. ҁࣴزᡉҢஒSVM ୔ϩၭӦख़ߎឦଯԦࢉወ༈୔ϐޜ໔ᆶឦ܄ၗ਑Ǵ঺᠄ݞ οࢬୱǵπቷǵπ཰୔ޑޜ໔ϩѲǴว᝺ӭೀଯԦࢉወ༈ϐ୔ୱឦߚଽว൑֟Ǵ ӭՏܭπቷǵπ཰୔ڬᜐǴЪԦࢉᒿ෤ၰࢬ୏໺ᒡᘉණǴࡺ SVM ᄽᆉݤ܌Ⴤϐ Ԧࢉወ༈კڀଯӝ౛܄ǴёշԖᜢൂՏղញᆶ،฼Ƕ

ୖԵЎ᝘

ԃનድǴ2011Ƕъᅱ࿎Ѝ࡭ӛໆᐒᆕॊǴࠄ٧εᏢीᆉᐒࣽᏢᆶמೌسǶ ՉࡹଣǴ2011Ƕύ๮҇୯Չ཰኱ྗϩᜪȐಃ9ԛঅुȑǴՉࡹଣЬीೀጓӑǶ Չࡹଣၭ཰ہ঩཮Ǵ2014Ƕឲ෸Н፦ᅱෳፓࢗϷמೌᇶᏤǴ103 ԃࡋၭ཰ว৖ी

ฝǶ

Չࡹଣᕉნߥៈ࿿Ǵ1998Ƕঅ҅ȨӦय़НᡏϩᜪϷН፦኱ྗȩǴᕉ࿿Нӷಃ0039159 ဦзǶ

Չࡹଣᕉნߥៈ࿿Ǵ1999ǶวթȨβᝆϷӦΠНԦࢉ᏾ݯݤϦѲࡼՉࡕၸ෠ਔය

୺ՉाᗺȩǴᕉ࿿ቲӷಃ0024062ဦзǶ

Չࡹଣᕉნߥៈ࿿Ǵ2011Ƕঅ҅ȨβᝆԦࢉ᏾ݯȩǴᕉ࿿βӷಃ0990119003ဦзǶ

׵٥݊ǵ஭ӂӓǵ຤ӹआǵЦࡿǴ2009Ƕϣఘᛥࡰኧຑሽݤޑঅ҅ϷځᔈҔǴН

ၗྍߥៈǴಃ6යǶ

ڬࡌԋǴ1990Ƕᆵ᡼ݞοН፦ࡰኧϐࡌҥǴ୯ҥԋфεᏢᕉნπำࣴز܌ᅺγፕ ЎǶ

ࢫऍذǴ2013Ƕᆵ᡼ၭӦख़ߎឦଯԦࢉወ༈୔ୱᑔᒧБݤϐ௖૸Ǵᆵ᡼εᏢғނ ᕉნس಍πำᏢسᅺγፕЎǶ

ࢫऍذǵᎄԭՙǵ৪຦ཥǵ஭൧୯Ǵ2012Ƕϣఘᛥࡰ኱ݤຑ݋఩НݞН፦Ǵ101 ԃࡋၭ཰πำࣴ૸཮Ƕ

ਲ༜ᑜࡹ۬Ǵ1998Ƕਲ༜ᑜ಍ीाំǴਲ༜Ǻਲ༜ᑜࡹ۬рހǶ

શྨ׊Ǵ2010Ƕ௖૸ȨНྍߥៈ୔ݞοԦࢉН፦ࡰ኱ȩϐᔈҔୢᚒǴᕉߥᙁૻǴ

ಃ6යǶ

஭൧୯Ǵ1994ǶճҔӦ౛ၗૻس಍ܭβᝆԦࢉ฻ભ୔ϩᆶወ༈ႣෳǴՉࡹଣ୯ৎ

ࣽᏢہ঩཮஑ᚒࣴزीฝԋ݀ൔ֋ǴಃѤകǴ17-22Ƕ

஭൧୯Ǵ2002Ƕᆵ᡼Ӧ୔βᝆԦࢉ౜ݩᆶ᏾ݯࡹ฼ϩ݋Ǵ଄იݤΓ୯ৎࡹ฼ࣴز

஭൧୯Ǵ2010Ƕਲ༜ၭӦԦࢉϐᕉნၗ਑ᇆ໣ᆶԦࢉᜢᖄ܄ϩ݋ीฝൔ֋ਜǴՉ

ࡹଣᕉߥ࿿ीฝǴEPA-100-GA101-03-A209Ƕ

஭൧୯Ǵ2010ǶᄆϯၭӦԦࢉϐᕉნၗ਑ᇆ໣ᆶԦࢉᜢᖄ܄ϩ݋ीฝǴՉࡹଣᕉ ߥ࿿ीฝǴEPA-100-GA103-02-D054Ƕ

஭൧୯Ǵ2011Ƕӄ୯ख़ߎឦଯԦࢉወ༈ၭӦϐᆅڋϷፓࢗीฝൔ֋ਜǴՉࡹଣᕉ ߥ࿿ीฝǴEPA-99-G101-03-A181Ƕ

஭൧୯ǵ݅ျரǴ2000ǶӦ౛಍ीኳᔕᆶ՗ीݤຑ՗βᝆख़ߎឦԦࢉጄൎǴՉࡹ

ଣ୯ৎࣽᏢہ঩཮஑ᚒࣴزीฝԋ݀ൔ֋ǴNSC89-2621-B-002-004Ƕ

஭൧୯ǵᆅ҉དǵᎄԭՙ฻Ǵ2012Ƕӄ୯ख़ߎឦଯԦࢉወ༈ၭӦϐᆅڋϷፓࢗी

ฝǴՉࡹଣᕉߥ࿿ीฝǴEPA-99-G101-03-A181Ƕ

஭൧୯ǵᎄԭՙǴ2012Ƕ101 ԃࡋཥчǵѠύǵଯ໢ၭӦԦࢉϐᕉნၗ໣ᆶԦࢉ

ᜢᖄ܄ϩ݋ीฝǴՉࡹଣᕉߥ࿿ीฝǴEPA-101-GA11-03-D167Ƕ

໱ذীǵص௵דǵᒽࡘᒍǵഋ؊໦Ǵ2011ǶᡧԀόӕભϦၡݮጕၭҖβᝆख़ߎ㳩 ԦࢉຑሽࣴزǴԖՅߎឦࣽᏢᆶπำǴಃ2යಃ1ڔǴ68-73Ƕ

࿶ᔮ೽Нճ࿿Ǵ2008ǶӦ౛ၗૻঊᓯύЈǶ

ቅ़։ǵ෯ቼཥǵқਁ๮ǵ஭ذ࣓ǵ஭ߥ๮Ǵ2009Ƕ୷ܭӦ፦ಕᑈᆶϣఘᛥࡰኧ ޑહӦख़ߎឦԦࢉࣴزǴύ୯ၭᏢ೯ൔǴಃ25යಃ20ڔǴ174-178Ƕ

ቅঅሎǵ࠴ٵဖǵ৪຦ཥǵᎄԭՙǵ஭൧୯Ǵ2014Ƕϣఘᛥࡰ኱ຑ݋ݞοНᡏН

፦ϩᜪϐၲ኱ำࡋǴ103ԃࡋၭ཰πำࣴ૸཮Ƕ

Boser, B., Guyon, M., Vapnik, V., 1992. A training algorithm for optimal margin classifiers. 5th Annual ACM Workshop on COLT, pp. 144-152.

Chang, C., Lin, C., 2011. LIBSVM: A library for support vector machines. ACM Transactions on Intelligent Systems and Technology 2(3), 27:1-27:27.

Chen, Z.S., Lee, D.Y., 1995. Heavy metals contents of representative agricultural soils in Taiwan. Journal of Chinese Institute of Environmental Engineering 5(3), 205-211.

Cheng, B.Y., Fang, W.T., Shyu, G.S., Chang, T.K., 2012. Distribution of heavy metals in the sediments of agricultural fields adjacent to urban areas Central Taiwan.

Paddy and Water Environment 11, 343-351.

Chong, X., Fuchu, D., Xiwei, X., Yuan, H., 2012. GIS-based support vector machine modeling of earthquake-triggered landslide susceptibility in the Jianjiang River watershed, China. Geomorphology 145-146, 70-80.

Deng, H.G., Gu, T.F., Li, M.H., Deng, X., 2012. Comprehensive Assessment Model on Heavy Metal Pollution in Soil. International Journal of Electrochemical Science 7, 5286-5296.

Ding, Y., Song, X., Zen, Y., 2008. Forecasting financial condition of Chinese listed companies based on support vector machine. Expert Systems with Applications 34, 3081-3089.

Hern੺ndez-S੺nchez, C., Luis, G., I.Moreno, Came੺n, A., 2012. Differentiation of mangoes (Magnifera indica L.) conventional and organically cultivated according to their mineral content by using support vector machines. Talanta 97, 325-330.

Hsu, Z.Y., Su, S.W., Lai, H.Y., Guo, H.Y., Chen, T.C., Chen, Z.S., 2010. Remediation techniques and heavy metal uptake by different rice varieties in metal-contaminated soils of Taiwan: New aspects for food safety regulation and sustainable agriculture. Soil Science and Plant Nutrition 56(1), 31 - 52.

Huang, S.W., Jin, J.Y., 2008. Status of heavy metals in agricultural soils as affected by different patterns of land use. Environmental Monitoring and Assessment 139,

Huang, Z., Chen, H., Hsu, C.J., 2004. Credit rating analysis with support vector machine and neural networks: A market comparative study. Decision Support Systems 37, 543-558.

Keerthi, S.S., Lin, C.J., 2003. Asymptotic behaviors of support vector machines with Gaussian kernel. Neural Computation 15(7), 1667-1689.

Kohavi, R., Provost, F., 1998. Glossary of terms - Special Issue on Applications of Machine Learning and the Knowledge Discovery Process. Machine Learning 30(2-3), 271-274.

Lin, H.-T., Lin, C.J., 2003. A study on sigmoid kernels for SVM and the training of non-PSD kernels by SMO-type methods. Technical Report, Department of Computer Science, National Taiwan University.

Maysam, A., Gholam-Hossai,n N., Abbas, B., 2012. Support vector machine for multi-classi¿cation of mineral prospectivity areas. Computers & Geosciences 46, 272-283.

Mikhail, K., Nicolas, G., 1999. Environmental Spatial Data Classification with Support Vector Machines. IDIAP-RR-99-07.

Miloš, K., Branislav, B., Boško, G., 2010. Soil type classification and estimation of soil properties using support vector machines. Geoderma 154, 340-347.

Nemerow, N.L., 1974. Scientific Stream Pollution Analysis. McGRAW-WILL BOOK COMPANY, New York, pp. 1-358.

Nemerow, N.L., 1985. Stream, Lake, Estuary and Ocean Pollution. Van Nostrand Reinhold Publishing Co., New York, pp. 303-309.

Pandey, G., Bin, Z., Le, J., 2013. Predicting submicron air pollution indicators: a

Powers, David, M.W., 2011. Evaluation: From precision, recall and F-measure to ROC, informedness, markedness & correlation. Journal of Machine Learning Technologies 2(1), 37-63.

Taner San, B., 2014. An evaluation of SVM using polygon-based random sampling in landslide susceptibility mapping: The Candir catchment area (western Antalya, Turkey). International Journal of Applied Earth Observation and Geoinformation 26, 399-412.

Tang, T., Zhai, Y., Huang, K., 2011. Water Quality Analysis and Recommendations through Comprehensive Pollution Index Method. Management science and engineering 5(2), 95-100.

Tay, F.E.H., Cao, L., 2001. Application of support vector machinesin financial time series forecasting. Omega 29, 309-317.

Vapnik, V., 1995. Support-vector networks. Machine Learning 20, 273-297.

Yang, Q., Lia X., Shi X., 2008. Cellular automata for simulating land use changes based on support vector machines. Computers & Geosciences 34, 592-602.

Yao, P.-H., Chang, T.-K., Shyu, G.-S., Cheng, B.-Y., Kuan, A., Chen, S.-D., Ho, J.-R., 2014. The course of protecting agricultural land from heavy metal pollution in Taiwan. 2014 International Conference on Remediation and Management of Soil and Groundwater Contaminated Sites, Taipei, Taiwan.

Yao, X., Tham, L.G., Dai, F.C., 2008. Landslide susceptibility mapping based on Support Vector Machine: A case study on natural slopes of Hong Kong, China.

Geomorphology 101, 572-582.

Zuo, R., Carranza E.J., 2011. Support vector machine: A tool for mapping mineral prospectivity. Computers & Geosciences 37, 1967-1975.