ಛ௃ྔͷݕ౼

࡞ۀ಺༰ਪఆͷਫ਼౓Λ޲্ͤ͞Δʹ͸ɼσʔλܭଌͱಛ௃நग़ͷਫ਼౓ɼ࡞ۀͷϧʔϧ Խɼಛ௃ྔͱ࡞ۀͷରԠ෇͚͕ͦΕͧΕ͏·͘ػೳ͢Δ͜ͱ͕ॏཁͰ͋Δɽಛ௃ྔͷ؍఺

͔ΒΈΔͱɼܭଌσʔλ͔Βਖ਼֬ʹநग़Ͱ͖ɼ͔ͭಛఆͷ࡞ۀͱͷରԠ͕෇͚ΒΕΔ͔Ͳ

͏͔͕ݤͱͳΔɽઌߦࣄྫͰ͸ɼՃ଎౓ηϯα΍ಈը૾ͷ৘ใΛ࡞ۀ಺༰ਪఆʹ׆༻ͯ͠

͍ͨɽث۩ʹηϯαΛ͚ͭΔ͜ͱͰɼͦͷث۩Λ࢖͏ͱ͍͏ಈ࡞Λݕग़͢Δ͜ͱ͕Ͱ͖

Δɽಈը૾͔Β͸ɼͦͷ৔ॴʹ͍ͯɼͲͷΑ͏ͳಈ࡞Λ͔ͨ͠ݕग़͢Δ͜ͱ͕Ͱ͖Δɽͦ

ΕΒͷಈ࡞ͱਪఆ͍ͨ͠࡞ۀ಺༰͕ඥ෇͚ΒΕΕ͹ྑ͍ɽ

ҰํɼຊݚڀͰର৅ͱ͢Δݱ৔͸ɼ࡞ۀ಺༰͕ෳࡶͰ͋Γɼ࡞ۀͷϧʔϧԽ͕೉͍͠ɽ

ྫ͑͹ɼʮѫࡰɾҊ಺ʯͱ͍͏࡞ۀ಺༰͸ɼैۀһ͕ސ٬ʹѫࡰΛͯ͠ɼر๬ͷ੮·ͰҊ಺

͠ɼҊ಺ઌ͔Β཭ΕΔ·Ͱͷಈ࡞Λࢦ͢ɽೖΓޱ෇͔ۙΒ٬੮΁ͷҠಈ͕ଟ͍ɼʮ͍Βͬ

͠Ό͍·ͤʯͳͲҊ಺ʹؔ͢ΔΩʔϫʔυΛൃ࿩͢Δɼͦͷ਺෼ޙʹ஫จ͕ೖΔͷͰձܭ σʔλʹه࿥͕͋Δɼͱ͍ͬͨࡉ͔ͳϧʔϧ͸গͳ͔Βͣ͋Δ͕ɼࣅͨΑ͏ͳϧʔϧΛ

࣋ͬͨ࡞ۀ͕ଞʹ΋͋ΔͨΊɼͲͷΑ͏ʹ૊Έ߹Θͤͯ࡞ۀ಺༰ͱରԠ෇͚Δ͔͕՝୊ͱ ͳΔɽ

ຊݚڀͰѻ͏σʔλ͸ɼैۀһͷҐஔɼํҐɼಈ࡞ͳͲΛදݱ͢Δηϯασʔλɼैۀ һͷൃ࿩΍पғͷԻΛه࿥͢ΔԻ੠σʔλɼैۀһ͕ೖྗ͢Δձܭσʔλͷ3^छྨͰ͋

Δɽ͍ͣΕ΋ɼ͋Δఔ౓ঢ়گ΍࡞ۀΛݶఆ͢Δ΋ͷͰ͸͋Δ͕ɼܾఆతͱݴ͑Δ΋ͷ͸গ ͳ͍ɽ·ͨɼਪఆ͢Δ࡞ۀ͸جຊಈ࡞Λෳ਺૊Έ߹Θͤͨෳࡶͳಈ࡞Ͱ͋Γɼ1ճͷ࡞ۀ ͷ͕۠ؒ௕͍ͨΊɼಉ͡࡞ۀ۠ؒͰಛ௃ྔ͕มԽ͢Δ৔߹΋ଟ͍ɽͦ͜Ͱɼ༷ʑͳಛ௃ྔ

Λ૊Έ߹Θͤͯ࡞ۀ಺༰ਪఆͷਫ਼౓ΛߴΊΔͨΊɼҎԼͷΑ͏ͳಛ௃ྔΛݕ౼ͨ͠ɽ

ਤ5.3 ࡞ۀ಺༰ʹΑΔൃ࿩཰ͷҧ͍ɽ

5.3.1 Ի੠σʔλʹجͮ͘ಛ௃ྔ

ຊݚڀͰ͸ɼԻ੠σʔλʹجͮ͘ಛ௃ྔͱͯ͠ୈ3ষͷൃ࿩۠ؒݕग़݁Ռͱɼୈ4ষ ͷ࿩ऀΫϥε෼ྨ݁ՌΛ༻͍Δɽද5.2ʹɼͦͷ֓ཁͱ࣍ݩ਺Λࣔ͢ɽൃ࿩཰ʢSpeech Rateʣͱ͸ɼҰఆ۠ؒʹ͓͚ΔԻ੠σʔλதͷൃ࿩۠ؒͷׂ߹Ͱ͋Δɽྫ͑͹ɼҰఆ۠ؒ

Λ5^{ඵͱͨ͠ͱ͖ͦͷ͏ͪ}2ඵؒൃ࿩͍ͯ͠Ε͹ɼൃ࿩཰͸40%^Ͱ͋Δɽਤ5.3^͸ɼද 5.1Ͱ֤ࣔͨ͠࡞ۀΛ͍ͯͨ۠ؒ͠ͷൃ࿩཰Ͱ͋Δɽͱ͘ʹ઀٬࡞ۀʢ࡞ۀNo.1ɼ3ʙ6ʣ Ͱ͸ैۀһͷൃ࿩཰͕3ׂΛ௒͑ΔҰํɼඇ઀٬࡞ۀʢ࡞ۀNo.7ʙ8ʣͰ͸ैۀһͷൃ࿩

཰͸ଞͷͦΕͱൺ΂ͯ௿͘ɼ઀٬ʗඇ઀٬ͷ࡞ۀΛ෼ྨ͢Δ্Ͱ͸ൃ࿩཰͸༗༻ͳಛ௃ྔ

ͱߟ͑ΒΕΔɽ

ྔࢠԽൃ࿩཰ʢQuantitative Speech Rateʣͱ͸ɼൃ࿩཰Λ֦ுͨ͠΋ͷͰɼൃ࿩۠ؒ

ݕग़ͷࣝผ෦ͷ͖͍͠஋Λෳ਺༻ҙ͠ɼ֤͖͍͠஋Λ௒͑ͨൃ࿩۠ؒͷׂ߹Λࢦ͢ɽ͖͠

͍஋Λෳ਺༻͍Δཧ༝ͱͯ͠ɼ༷ʑͳύϫʔ΍໌ྎੑΛ࣋ͭൃ࿩Λ۠ผ͢Δ͜ͱ͕ڍ͛Β ΕΔɽൃ࿩ʹ͸ɼѫࡰ΍ۀ຿࿈བྷͷΑ͏ʹ཭Εͨ૬खʹରͯ͠େ͖ͳύϫʔͰൃ੠͢Δ΋

ͷ͔Βɼ٬੮Ͱͷ઀٬΍ैۀһಉ࢜ͷࡶஊͳͲ໨ͷલͷ૬खʹରͯ͠খ͞ͳύϫʔͰൃ੠

͢Δ΋ͷ·Ͱ͋Δɽ͜ΕΒͷൃ੠๏Λ۠ผͯ͠ݕग़Ͱ͖Ε͹ɼൃ੠๏ʹԠͨ͡࡞ۀ಺༰ͷ

෼ྨʹد༩͢Δͱߟ͑ΒΕΔɽ͜ΕΒ͸ɼൃ࿩۠ؒݕग़ͷಛ௃நग़෦ͰಘΒΕΔಛ௃ྔʹ

͕ࠩΈΒΕΔͨΊɼ͖͍͠஋ॲཧʹΑͬͯ۠ผ͢Δɽൃ࿩۠ؒݕग़ʹSVM΍GMMͳ Ͳͷڭࢣ͋Γख๏Λద༻ͨ͠৔߹ɼൃ࿩ͷ༧ଌ֬཰Λ͖͍͠஋ॲཧ͢Δ͜ͱͰɼ΄΅ಉٛ

ͱͳΔɽද5.3ʹɼ͋Δ۠ؒͷൃ࿩཰ͱྔࢠԽൃ࿩཰ͷྫΛࣔ͢ɽຊݚڀͰ͸ɼ͖͍͠஋

Λ8छྨ༻͍ͨɽൃ࿩۠ؒݕग़ͷֶशσʔλʹ͓͍ͯ100%^{ൃ࿩ͱͳΔ͖͍͠஋ʢ}th8ʣ Λ࠷େͱ͠ɼ50%^{ൃ࿩ʢ࢒Γͷ}50%͸ඇൃ࿩ʣͱͳΔ͖͍͠஋Λ࠷௿ͷ͖͍͠஋ʢth1ʣ ͱͨ͠ɽͦͯ͠ɼͦͷؒͰ౳ִؒʹ8ஈ֊ͷ͖͍͠஋Λઃఆͨ͠ɽ

ද5.2 Ի੠σʔλʹجͮ͘ಛ௃ྔҰཡʢ17^࣍ݩʣɽ

ಛ௃ྔ໊ ࣍ݩ਺ ֓ཁ

(i) ྔࢠԽൃ࿩཰ 8 ൃ࿩۠ؒݕग़ͷ͖͍͠஋Λม͑ͯٻΊͨෳ਺ͷൃ࿩۠ؒ

(ii) ൃ࿩ͷ༗ແ 9 ϚΠΫ૷ணऀɼଞͷैۀһɼސ٬ͷ3࿩ऀΫϥεʹ͍ͭͯ

3छྨͷ۠ؒͰͷൃ࿩ͷ༗ແ

ද5.3 ͋Δ࡞ۀ۠ؒͷൃ࿩཰ͱྔࢠԽൃ࿩཰ɽ

#frame

ൃ࿩཰[%] ^{ྔࢠԽൃ࿩཰}[%]

start end th1 th2 th3 th4 th5 th6 th7 th8

0 99 0 10 0 0 0 0 0 0 0

100 199 54 40 30 20 10 10 10 10 0

200 299 100 80 70 60 60 60 50 20 10

300 399 100 100 90 80 60 60 50 20 10

400 499 12 30 0 0 0 0 0 0 0

ൃ࿩ͷ༗ແͱ͸ɼ͋Δ۠ؒͰൃ࿩͕͔͋ͬͨͲ͏͔Λ2஋Խͨ͠ಛ௃ྔͰ͋Δɽൃ࿩͕

͋Ε͹1^{ɼൃ࿩͕ͳ͚Ε͹}0ͱ͢Δɽ͋Δ۠ؒͱ͸ɼਤ5.4^{ʹࣔ͢Α͏ͳ}3^{छྨͷ۠ؒͱ}

͢ΔɽT ͸࡞ۀ಺༰ਪఆ͢Δ࠷খͷ۠ؒ௕Ͱ͋ΓɼU ^͸T ^ΛؚΈT ^{ΑΓ΋௕͍۠ؒ௕Ͱ}

͋Δɽ͢ͳΘͪɼT^{ඵͷ࡞ۀ۠ؒɼ}Uඵલ͔Βͦͷ࡞ۀ۠ؒ·Ͱɼͦͷ࡞ۀ͔۠ؒΒU^ඵ ޙ·Ͱͷ3ͭͷ۠ؒʹ͓͚Δൃ࿩ͷ༗ແΛಛ௃ྔͱ͢Δɽൃ࿩཰ͱҧͬͯൃ࿩ͷ௕͞Λߟ

ྀͤͣɼΔT ^{Ҏ্ͷൃ࿩͕͋Ε͹}1ͱ͢Δɽ͜Ε͸ɼಥൃతͳࡶԻʹΑΔൃ࿩ͷޡݕग़͕

ߟ͑ΒΕΔͨΊͰ͋Δɽൃ࿩͸1ճͷ࡞ۀͱൺ΂͕ͯ࣌ؒ୹͘ɼൃ࿩۠ؒݕग़΍࿩ऀΫϥ ε෼ྨʹ͓͚Δ෼ੳ૭͝ͱʹ൑ఆ͢Δͱɼͦͷଟ͘͸0ʹͳͬͯ͠·͏ͨΊɼ࡞ۀ಺༰ਪ ఆͰ͸0ʹͳΔӨڹΛ཈͑ΔͨΊɼൃ࿩ͷଘࡏ͢Δ۠ؒΛશͯ1ͱ͢Δɽ·ͨɼҰ෦ͷ࡞

ۀ಺༰ʹ͍ͭͯ͸ɼൃ࿩͕Ҏલʹ͋ͬͨɼ͋Δ͍͸ൃ࿩͕ͦͷޙʹ͋Δɼͱ͍͏৘ใ΋ॏ

ཁͰ͋Δɽྫ͑͹ɼҠಈɾӡൖͱ͍͏࡞ۀͷ৔߹ɼ࡞ۀલͱ࡞ۀޙʹൃ࿩͕͘Δ৔߹͕ଟ

͍ɽຊݚڀͰ͸ɼ࿩ऀΫϥε෼ྨʹΑͬͯಘΒΕͨ3ͭͷ࿩ऀΫϥεʢϚΠΫΛ૷ணͨ͠

ैۀһɼଞͷैۀһɼސ٬ʣʹ͍ͭͯɼͦΕͧΕ3छྨͷ۠ؒͰൃ࿩ͷ༗ແΛ൑ఆ͠ɼ߹

ܭ9࣍ݩͷ৘ใΛಛ௃ྔͱͨ͠ɽ·ͨɼT^ɼU^ɼΔT ^ΛͦΕͧΕ5^ɼ60^ɼ1^ͱͨ͠ɽ

5.3.2 ηϯασʔλʹجͮ͘ಛ௃ྔ

ैۀһʹ͸ࠊʹখܕͷηϯαϞδϡʔϧΛ૷ண͠ɼาߦऀσουϨίχϯάʢPedestrian

Dead-ReckoningʀPDRʣʹج͍ͮͯैۀһͷ૬ରଌҐΛߦͬͨɽ·ͨɼళฮʹઃஔͨ͠

ਤ5.4 ^{ൃ࿩ͷ༗ແΛௐࠪ͢Δ}3^{ͭͷ࣌ؒଳɽ}

RFIDλά΍ϏσΦΧϝϥͳͲͷଌҐΠϯϑϥʹΑΔैۀһͷઈରଌҐΛߦͬͨɽͦ͠

ͯɼ͜ΕΒΛ౷߹͢Δηϯασʔλ౷߹ʢSensor Data FusionʀSDFʣٕज़ʹΑΓɼैۀ һͷҐஔɼํҐɼಈ࡞ͳͲΛਪఆͨ͠[21]^ɽͦΕͧΕ1ඵ͝ͱʹਪఆ͠ɼҐஔʹ͍ͭͯ͸

ͦͷ࣌ࠁͷैۀһͷX^࠲ඪɼY^࠲ඪɼZ࠲ඪʢߴ͞ʣͷ৘ใɼํҐʹ͍ͭͯ͸ͦͷ࣌ࠁͷ

ैۀһͷํҐʢϥδΞϯ஋ʣɼಈ࡞ʹ͍ͭͯ͸Ճ଎౓৘ใΛ͖͍͠஋ॲཧ͠ɼͦͷ࣌ࠁʹ্

Լӡಈ΍าߦಈ࡞͕͔͋ͬͨͲ͏͔Λ൑ఆͯ͠2஋Խͨ͠ɽද5.4ʹɼ͜ΕΒͷ৘ใ͔Β ܭࢉ͞ΕΔɼ࡞ۀਪఆʹ༻͍Δಛ௃ྔͷҰཡΛࣔ͢ɽ·ͨɼ֤ಛ௃ྔͷઆ໌Λޙड़͢Δɽ

ද5.4 ηϯασʔλʹجͮ͘ಛ௃ྔҰཡʢ19^࣍ݩʣɽ

ΧςΰϦ ಛ௃ྔ໊ ࣍ݩ਺

Ґஔ (i) ^{ΤϦΞ଺ࡏ཰} 8 (ii) ^{ΤϦΞ௨ա਺} 2

ಈ࡞ (iii) ^{า਺ɾาߦಈ࡞཰} 2

(iv) ্Լಈ࡞཰ 1

ํҐ (v) ֤ํҐΛ޲ׂ͍ͨ࣌ؒ߹ 4 (vi) ํҐͷมԽ਺ 2

(i)^{ΤϦΞ଺ࡏ཰}

ैۀһͷҐஔ৘ใͱͯ͠ɼX^࠲ඪɼY^࠲ඪ͕1ඵ͝ͱʹਪఆ͞Ε͍ͯΔɽ͔ͦ͜ΒɼҰ ఆ࣌ؒ͝ͱʹΤϦΞ଺ࡏ཰ɼΤϦΞ௨ա਺ͳͲΛٻΊΔɽΤϦΞ଺ࡏ཰ͱ͸ɼళฮΛ͍͘

͔ͭͷΧςΰϦʢ٬ࣨɼਥ๪ͳͲʣͷΤϦΞʹ෼ׂ͠ɼͦͷΤϦΞͷൣғ಺ʹ଺ࡏ͍ͯ͠

Δ࣌ؒͷׂ߹Λද͢ɽళฮͷ෼ׂਤΛਤ5.5ʹࣔ͢ɽ·ͨɼΤϦΞͷ֤ΧςΰϦͷఆٛΛ ද5.5ʹࣔ͢ɽਤ5.5ͷΤϦΞͷ৭͸ΧςΰϦΛද͢ɽձܭ࡞ۀͰ͋Ε͹ɼϨδ΍ళฮͷ

ೖޱ෇ۙʹ଺ࡏ͢Δ࣌ؒͷׂ߹͕௕͘ɼϨδ଺ࡏ཰΍ೖޱ଺ࡏ཰ͷ஋͕େ͖͘ͳΔͱߟ͑

ΒΕΔɽಉ༷ʹɼҠಈɾӡൖ࡞ۀͰ͸ɼ௨࿏଺ࡏ཰ͷ஋͕େ͖͘ͳΔͱߟ͑ΒΕΔɽ5^ඵ

୯ҐͰ࡞ۀΛਪఆ͢Δ৔߹ɼ5^ඵؒͷ͏ͪ4^{ඵϨδʹ଺ࡏ͠ɼ࢒Γ}1^{ඵ௨࿏ʹ଺ࡏ͍ͯ͠}

Ε͹ɼϨδ଺ࡏ཰͸80%ɼ௨࿏଺ࡏ཰͸20%ɼͦͷଞͷΤϦΞͷ଺ࡏ཰͸0%ͱ͢Δɽ (ii)^{ΤϦΞ௨ա਺}

ΤϦΞ௨ա਺͸ɼैۀһ͕Ұఆ࣌ؒ಺ʹΤϦΞͱΤϦΞͷڥքΛ௨աͨ͠ճ਺Ͱ͋Δɽ

ྫ͑͹ɼ٬͔ࣨΒ௨࿏ɼ·ͨ͸௨࿏͔Β٬ࣨ΁Ҡಈͨ͠৔߹ɼ1ճͱΧ΢ϯτ͢Δɽ٬ࣨ

͔Β௨࿏ɼ͔ͭ௨࿏͔Β٬ࣨ΁Ҡಈͨ͠৔߹͸2ճͱΧ΢ϯτ͢ΔɽαϒΤϦΞ௨ա਺

͸ɼΤϦΞΛ͞Βʹ෼ׂͨ͠αϒΤϦΞʹରͯ͠ɼαϒΤϦΞؒͷ௨աճ਺ΛΧ΢ϯτ͢

Δɽਤ5.5ͷ֤αϒΤϦΞͷ൪߸͸ɼද5.5ͷαϒΤϦΞͷID ͱରԠ͢Δɽ࣮ݧͰ͸ɼ ΤϦΞ௨ա਺ͱαϒΤϦΞ௨ա਺ʢܭ2࣍ݩʣͷಛ௃ྔΛ࢖༻ͨ͠ɽ͜ͷಛ௃ྔ͸ɼҠ ಈɾӡൖ࡞ۀ΍ɼѫࡰɾҊ಺ͱ͍ͬͨɼෳ਺ͷΤϦΞΛҠಈ͠ͳ͕ΒߦΘΕΔ࡞ۀͷݕग़ ʹد༩͢Δͱߟ͑ΒΕΔɽ

(iii)^{า਺ɾาߦಈ࡞཰ɼ}(iv)^{্Լಈ࡞཰}

ηϯασʔλʹ͸ɼैۀһͷՃ଎౓σʔλ΍ɼา਺ܭͷσʔλ΋ؚ·ΕΔɽͦ͜Ͱɼै

ۀһͷಈ࡞৘ใͱͯ͠ɼैۀһͷา਺ɼาߦಈ࡞ɼ্Լಈ࡞ʹؔ͢Δಛ௃ྔΛٻΊͨɽ

ैۀһͷา਺͸ɼา਺ܭͷσʔλΛ΋ͱʹҰఆ۠ؒͷา਺ΛٻΊΔɽาߦಈ࡞͸ɼา਺

ܭͷ࣌ࠁͷσʔλΛ΋ͱʹɼ1ඵ͝ͱʹาߦಈ࡞ͷ༗ແΛݕग़͠ɼ୯Ґ࣌ؒͰूܭͯ͠ಛ

௃ྔԽ͢Δɽྫ͑͹ɼ5ඵؒͷ͏ͪ4ඵาߦಈ࡞Λ͍ͯ͠Δ৔߹͸ɼಛ௃ྔͷ஋͸80%ͱ ͳΔɽ্Լಈ࡞ʹ͍ͭͯ͸ɼՃ଎౓σʔλͷZ࣠ํ޲ͷ஋͕ᮢ஋Λ௒͑Δ͔Ͳ͏͔Λ1ඵ

͝ͱʹ൑ఆ͠ɼ্Լಈ࡞Λ൑ఆ͢Δɽͦͯ͠୯Ґ࣌ؒͷ͏্ͪԼಈ࡞͕ݕग़͞Εͨ࣌ؒͷ

ׂ߹Λಛ௃ྔͱ͢Δɽ

ͦͷ৔Ͱ੩ࢭͯ͠ߦ͏ձܭ࡞ۀ΍஫จ࢕͍Ͱ͸ɼ͜ΕΒͷಛ௃ྔͱෛͷ૬͕ؔ͋Δͱߟ

͑ΒΕΔɽ·ͨɼ഑હ΍٬ࣨ४උɾย෇͚ͳͲɼςʔϒϧʹରͯ͠ߦ͏࡞ۀͰ͸ɼ্Լಈ

࡞ͷճ਺͕૿͑ΔͳͲͷಛ௃͕දΕΔͱߟ͑ΒΕΔɽ (v)֤ํҐΛ޲ׂ͍ͨ࣌ؒ߹

ηϯασʔλ͔Βਪఆͨ͠Ґஔ৘ใͷଞʹɼैۀһͷ޲͍͍ͯΔํ֯Λࣔ͢ํҐ৘ใ͕

1ඵ͝ͱʹಘΒΕΔɽͦ͜ͰɼํҐΛ4ͭʹ෼ׂ͠ɼͦΕͧΕͷํҐΛ޲͍͍ͯΔ࣌ؒͷ

ׂ߹Λಛ௃ྔͱͯ͠༻͍ͨɽ͜Ε͸ɼಛఆͷํҐͰ࡞ۀΛߦ͏Α͏ͳ৔߹ʹɼ༗ޮͳಛ௃

ྔͰ͋Δͱߟ͑ΒΕΔɽ (vi)^{ํҐͷมԽྔ}

ํҐͷมԽྔͱ͸ɼҰఆ۠ؒͰैۀһͷํҐ͕มԽ͔ͨ͠Ͳ͏͔ΛΧ΢ϯτͨ͠ಛ௃

ྔͰ͋Δɽʮ഑હʯʮ٬ࣨ४උɾย෇͚ʯͷΑ͏ʹํҐ͕Α͘มԽ͢ΔʢମΛճసͤ͞Δʣ

࡞ۀͱɼʮ஫จ࢕͍ʯʮҠಈɾӡൖʯͷΑ͏ʹ͋·ΓมԽ͠ͳ͍ʢಛఆͷํ޲Λ޲͖ଓ͚

Δʣ࡞ۀͰҧ͍͕දΕΔͱߟ͑ΒΕΔɽํҐͷมԽྔ͸NoDC^ʢNumber of Direction

Change^ʣͱTAoDC^ʢTotal Amount of DC^ʣͷ2^{छྨΛఆٛͨ͠ɽ}NoDC^͸ɼ1^ඵ͝

ͱʹํҐʢ90౓ͣͭ4෼ׂͨ͠ํҐʣ͕มԽͨ͠ճ਺ΛΧ΢ϯτ͠ɼҰఆ۠ؒͰճ਺Λ߹

ܭ͢ΔɽTAoDC͸ɼ45౓มԽ͢Δ͝ͱʹ1ճΧ΢ϯτ͠ɼҰఆ۠ؒͰճ਺Λ߹ܭ͢Δɽ

͋ΔैۀһͷํҐͷσʔλΛθ(x)ͱ͢Δͱɼ࣌ࠁt^ɼ۠ؒ௕T ^ʹ͓͚ΔNoDCɼTAoDC

͸ͦΕͧΕࣜ(5.1)^ɼࣜ(5.2)^{Ͱܭࢉ͞ΕΔɽʢ˞} x^͸x^{Λখ਺఺੾Γࣺͯʣ}

f(t) =

t+T−1 i=t

min

|θ(i+ 1)−θ(i)| π/2 ,1

(5.1)

f(t) =

t+T−1 i=t

|θ(i+ 1)−θ(i)|

π/4 (5.2)

ਤ5.5 ళฮͷΤϦΞ෼ׂਤʢ্ɿB1^ɼԼɿB2ɼਤதͷ൪߸͸αϒΤϦΞID^Λࣔ͢ʣɽ

ද5.5 ΤϦΞͱαϒΤϦΞͷఆٛɽ

ΤϦΞ໊ ਤ5.5^{தͷ৭ αϒΤϦΞ}ID(B1) ^αϒΤϦΞID(B2)

ೖޱ ˙(^փ৭) 1,2

Ϩδ ˙(౧৭) 3 46

ௐཧ৔ ˙(੨৭) 4ʢҰ෦লུʣ

֊ஈ ˙(ࢵ৭) 17 44,45

ύϯτϦʔ ˙(྘৭) 18 40ʙ43 αʔϏεΩϟϏωοτ ˙(^੺৭) 19,20 47

٬ࣨɾ٬੮ ˙(^ᒵ৭) 21^ʙ28 48^ʙ54 ௨࿏ ˙(ԫ৭) 29ʙ39 55ʙ62

5.3.3 ձܭσʔλʹجͮ͘ಛ௃ྔ

ձܭσʔλ͸ސ٬਺΍஫จʹؔ͢Δ৘ใͰ͋Δɽैۀһ͸ސ٬͔Β஫จΛऔΔͱ͖ɼձ ܭΛ͢Δͱ͖ʹɼҎԼͷ৘ใΛPOS୺຤ʹೖྗ͢Δɽ

• ఻ථ൪߸ɼ஫จछผʢ৽نʗ௥Ճʣɼ༧໿ͷ༗ແ

• ^{஫จςʔϒϧɼΤϦΞ}

• ٬਺ɼ٬૚ʢੑผͱ೥ྸ૚ʣ

• ^{஫จ୲౰ऀ}

• ^{஫จ࣌ࠁɼձܭ࣌ࠁ*1}

• ஫จϝχϡʔɼ୯Ձɼ਺ྔ

஫จ࣌ࠁɼձܭ࣌ࠁ͸෼୯ҐͰه࿥͞Ε͍ͯΔɽ٬૚ͷ೥ྸ૚͸20^୅ɼ40^୅ɼ60^୅ͷ3 छྨͰ͋Δɽ஫จ࢕͍΍ձܭͷࡍʹ͜ΕΒͷσʔλΛೖྗ͢ΔͨΊɼೖྗ࣌ࠁ͸஫จ࢕͍

΍ձܭ࡞ۀͱͷؔ࿈͕ڧ͍ɽ·ͨɼ٬਺΍஫จ਺ͳͲɼձܭσʔλͷ౷ܭ͔Βۀ຿ͷ๩͠

͞΍܏޲͕දΕΔ͜ͱ͕ظ଴͞ΕΔɽຊ࣮ݧͰ͸ɼձܭσʔλ͔Βද5.6ʹࣔ͢ಛ௃ྔΛ நग़ͯ͠ɼ࡞ۀ಺༰ਪఆʹ࢖༻ͨ͠ɽৄࡉ͸ޙड़͢Δɽ

ද5.6 ձܭσʔλʹجͮ͘ಛ௃ྔҰཡʢ12^࣍ݩʣɽ

ΧςΰϦ ಛ௃ྔ໊ ࣍ݩ਺

஫จ (i) ஫จ࢕͍ɾձܭ࡞ۀ࣌ؒଳ 4 (ii) ஫จൃੜ਺ 2

٬਺ (iii) ^{ೖళɾୀళऀ਺} 4

(iv) ଺ࡏ٬਺ɾάϧʔϓ਺ 2

*1഑હ࣌ࠁɼௐཧ։࢝ɾ׬ྃ࣌ࠁͳͲͷ߲໨΋͋Δ͕ɼೖྗͷෛ୲্ه࿥͞Ε͍ͯͳ͍ɽ

(i)஫จ࢕͍ɾձܭ࡞ۀ࣌ؒଳ

஫จ࢕͍ɾձܭ࡞ۀ࣌ؒଳͱ͸ɼձܭσʔλͷ஫จ࣌ࠁɼձܭ࣌ࠁΛج४ʹͨ͠ಛ௃ྔ

Ͱ͋Δɽ஫จ࣌ࠁ·ͨ͸ձܭ࣌ࠁΛt0ͱͨ͠ͱ͖ɼಛ௃ྔf(t)͸ࣜ(5.3)Ͱද͞ΕΔɽ U ͸ฏ׈Խ෯Ͱ͋Δɽ·ͨɼඍখ۠ؒΔtͷಛ௃ྔΛٻΊΔͱ͖͸ࣜ(5.4)^{ͷΑ͏ʹඍখ}

۠ؒͰf(t)^{ͷฏۉ஋ΛͱΔɽ}

f(t) =

⎧⎨

⎩

1 (−U ≤t−t0<0)

−^t−t_U⁰ + 1 (0≤t−t0< U) 0 (otherwise)

(5.3)

fΔt(t) = 1 Δt

Δt−1

i=0

f(t+i) (5.4)

t0ʹൃੜͨ͠1ͭͷ஫จɾձܭʹର͢Δؔ਺f(t)Λਤ5.6ʹࣔ͢ɽෳ਺ͷ஫จ·ͨ͸ձ ܭ͕2U ඵҎ಺ʹ࿈ଓͯ͠ൃੜ͢Δ৔߹ɼਤ5.7ͷΑ͏ʹݸʑͷ஫จ·ͨ͸ձܭʹؔ͢Δ

ؔ਺ͷ࠷େ஋ΛऔΔɽձܭσʔλͷ஫จ࣌ࠁɼձܭ࣌ࠁͷલʹ஫จ࢕͍΍ձܭͷ࡞ۀΛ͠

͍ͯΔՄೳੑ͕ߴ͍ͨΊɼ͜ͷಛ௃ྔ͸஫จ࢕͍΍ձܭͷ࡞ۀͱͦΕҎ֎ΛΑ͘෼཭͢Δ

͜ͱ͕ظ଴͞ΕΔɽ࣮ݧͰ͸ɼ࣌ؒ௕Uʢ୯Ґ͸ඵʣ͸ෳ਺ͷ஋Λ༻͍ɼ஫จʹؔͯ͠͸

U = 120ɼ180ɼ240ɼձܭʹؔͯ͠͸U = 120Ͱؔ਺f(t)Λੜ੒͠ɼܭ4࣍ݩͷಛ௃ྔ

ΛٻΊͨɽ

ਤ5.6 ஫จ࢕͍ɾձܭ࡞ۀ࣌ؒଳʹ͍ͭͯͷؔ਺ɽ

ਤ5.7 ஫จɾձܭ࣌ࠁ͕Ұఆ۠ؒʹ࿈ଓ͢Δ৔߹ɽ

(ii)^{஫จൃੜ਺}

஫จൃੜ਺ͱ͸ɼ͋Δ࣌ࠁt−M+ 1͔Β࣌ࠁt^·ͰͷM ෼ؒʹൃੜͨ͠஫จ਺Ͱ͋

Δɽ஫จൃੜ਺͕ଟ͍ͱ͖͸ʮ஫จ࢕͍ʯʮ഑હʯͳͲͷ઀٬࡞ۀΛ͢ΔՄೳੑ͕ߴ͘ɼ஫

จൃੜ਺͕গͳ͍ͱ͖͸ʮ٬ࣨ४උɾย෇͚ʯͳͲͷඇ઀٬࡞ۀΛ͢ΔՄೳੑ͕ߴ͍ɽ࣮

ݧͰ͸M = 1,5ͷͱ͖ͷ஫จൃੜ਺ʢܭ2࣍ݩʣΛ࢖༻ͨ͠ɽ഑હ࣌ࠁ͕෼͔Δ৔߹͸ɼ

഑હ࣌ࠁͱ஫จ࣌ࠁͷؒͷ஫จ਺ΛΧ΢ϯτ͢Δ͜ͱͰɼΑΓਖ਼֬ͳ஫จൃੜ਺͕ٻΊΒ ΕΔɽ

(iii)^{ೖళɾୀళऀ਺}

ೖళɾୀళऀ਺͸ɼ͋Δ࣌ࠁtͷ෇ۙͰͲΕ͚ͩೖళऀ·ͨ͸ୀళऀ͕͍͔ͨΛද͢ɽ

ೖళऀ਺͸CGin1(Customer Group in 1)^ͱCGin2^ͷ2^{छྨΛఆٛͨ͠ɽ}CGin1^͸

࣌ࠁt^͔Βt+M ͷؒʹൃੜͨ͠৽ن஫จʹର͢Δ٬਺Ͱ͋Δɽैۀһ͸஫จΛͱΔલ ʹʮѫࡰɾҊ಺ʯΛ͢ΔՄೳੑ͕ߴ͍ͨΊɼ࣌ࠁt^͔ΒM෼ޙͷ஫จʹண໨͢ΔɽCGin2

͸࣌ࠁt−M ^͔Βt+M ͷؒʹൃੜͨ͠৽ن஫จ਺Ͱ͋Δɽ͜Ε͸ɼ஫จ࣌ࠁͷ෇ۙͰ

͸ैۀһ͕ʮ஫จ࢕͍ʯΛ͢ΔՄೳੑ͕ߴ͍͜ͱʹΑΔɽୀళऀ਺΋ɼೖళऀ਺ͱಉ༷ʹ CGout1^ɼCGout2^{Λఆٛͨ͠ɽ}CGout1^͸ɼ࣌ࠁt−M^͔Βt^{ͷؒʹൃੜͨ͠ձܭʹ} ର͢Δ٬਺Ͱ͋Δɽैۀһ͸ձܭޙʹʮ٬ࣨ४උɾย෇͚ʯΛ͢ΔՄೳੑ͕ߴ͍ͨΊɼձ ܭ࣌ࠁt^ͷM ^{෼લ͔Β࣌ࠁ}t^{ͷձܭʹண໨͢Δɽ}CGou2͸࣌ࠁt−M^͔Βt+M ^ͷؒ

ʹൃੜͨ͠ձܭʹର͢Δ٬਺Ͱ͋Γɼ७ਮʹʮձܭʯ࡞ۀͷݕग़Λ໨తͱ͍ͯ͠Δɽೖళɾ

ୀళऀ਺ͷΧ΢ϯτʹ͋Δఔ౓ͷ۠ؒΛઃ͚͍ͯΔͷ͸ɼಛ௃ྔ͕εύʔεʹͳΔ͜ͱΛ

๷͙ͨΊͰ͋Δɽ͋Δ࣌ࠁtͷೖళɾୀళऀ਺͚ͩΛݟͨ৔߹ɼେ෦෼ͷ࣌ࠁ͸ೖళऀɾ

ୀళऀ͕0ͱͳΔɽ͔͠͠ɼސ٬͕ೖళɾୀళ͢Δͱͦͷલޙ਺෼ͷ࡞ۀʹӨڹΛ༩͑Δ

ͨΊɼલޙͷ۠ؒΛؚΊͨೖళɾୀళऀ਺ΛΧ΢ϯτͨ͠ɽ࣮ݧͰ͸ɼCGin1ɼCGout1

͸M = 10ɼCGin2ɼCGout2͸M = 3ͱ͠ɼೖళɾୀళऀ਺ͦΕͧΕ2࣍ݩɼܭ4࣍ ݩͷಛ௃ྔΛٻΊͨɽ

(iv)^{଺ࡏ٬਺ɾάϧʔϓ਺}

஫จ࣌ࠁ͔Βձܭ࣌ࠁͷؒ͸ɼͦͷ஫จͷސ٬͸ళʹ଺ࡏ͍ͯ͠ΔͱΈͳ͢͜ͱͰɼ஫

จ࣌ࠁͱձܭ࣌ࠁΛ΋ͱʹɼͦͷ࣌఺Ͱͷ଺ࡏ٬਺΍଺ࡏάϧʔϓ਺ΛٻΊΔ͜ͱ͕Ͱ͖

Δ*2ɽ଺ࡏ٬਺͕ଟ͍৔߹͸઀٬࡞ۀΛ͍ͯ͠ΔՄೳੑ͕ߴ͘ɼ଺ࡏ٬਺͕গͳ͍৔߹͸

ͦͷଞͷ࡞ۀΛ͢ΔՄೳੑ͕ߴ͍ͱߟ͑ΒΕΔɽ଺ࡏ٬਺fc(t)͸ɼʮ࣌ࠁt^{·Ͱʹൃੜ͠}

ͨ৽ن஫จͷྦྷੵ٬਺ʯ͔Βɼʮ࣌ࠁt·ͰʹձܭΛऴ͑ͨ৽ن஫จͷྦྷੵ٬਺ʯΛҾ͍

ͨ΋ͷͰ͋Δɽ࣌ࠁ͸෼୯ҐͷͨΊɼ࣌ࠁtʹؚ·ΕΔ۠ؒͷ஋͸શͯfc(t)^ͱͳΔɽ଺

ࡏάϧʔϓ਺΋ɼ଺ࡏ٬਺ͱಉ༷ʹٻΊΔɽܭ2࣍ݩͷಛ௃ྔͱ࣮ͯ͠ݧʹ࢖༻ͨ͠ɽ

*2࣮ࡍʹ͸ɼ஫จ࣌ࠁΑΓલʹސ٬͕ೖళ͢ΔͨΊɼ଺ࡏ٬਺͸ਖ਼֬Ͱ͸ͳ͍ɽ

5.4 ^{ࣝผثͷݕ౼}

5.4.1 Adaboost ͷجຊ

FreundΒʹΑͬͯఏҊ͞ΕͨAdaboost๏[42]͸ɼऑ෼ྨثΛ૊Έ߹Θͤͯڧ෼ྨث Λߏ੒͢ΔϒʔεςΟϯάͷҰछͰ͋Γɼύλʔϯೝࣝʹ͓͚Δֶशख๏ͷ1ͭͰ͋Δɽ

Adaboost๏ͷಛ௃͸ɼऑ෼ྨث͕ޡͬͨαϯϓϧʹର͢ΔॏΈΛେ͖͘͢Δߋ৽ଇʹ͋

Γɼऑ෼ྨثΛదԠతʹֶश͢Δ͜ͱͰɼਫ਼౓ͷ௿͍ʢϥϯμϜΑΓ͸ਫ਼౓͕ߴ͍ʣ෼ྨ

ثΛ׆͔ͯ͠ڧ෼ྨثΛੜ੒Ͱ͖Δɽ

Adaboost๏ͷֶशΞϧΰϦζϜ͸࣍ͷ௨ΓͰ͋Δɽm ^{ݸͷσʔληοτ} (x1, y1) ɼ· · ·,(xm, ym) ͕༩͑ΒΕͨͱ͢Δɽxi͸֤αϯϓϧͷ؍ଌ৴߸ɼyi (yi ∈ {−1,+1})

͸ڭࢣ৴߸Ͱ͋Δɽ·ͨɼh(x)Λ؍ଌ৴߸x^͔Β{−1,+1}Λग़ྗ͢Δࣝผؔ਺ͱ͢Δɽ

ֶशσʔλͷ֤αϯϓϧi^ͷॏΈΛDt(i)ͱ͓͘ɽt^{͸ֶशεςοϓ਺Ͱɼ}1≤t≤T ^ͱ

͢ΔɽT ͸Ұൠʹ͸ऑࣝผثͷݸ਺Ͱ͋Δɽ·ͣɼॏΈͷॳظ஋D1(i)^{ΛҎԼͷΑ͏ʹઃ}

ఆ͢Δɽ

D1(i) = 1

m (5.5)

ଓ͍ͯɼશͯͷऑࣝผثʹ͍ͭͯɼऑࣝผثj^ͷޡΓ౓j Λࣜ(5.6)Ͱܭࢉ͠ɼ࠷΋j

ͷখ͍͞ऑࣝผثͷࣝผؔ਺hj(x)Λͦͷֶशεςοϓt^{ʹ͓͚Δࣝผؔ਺}ht(x)ʹ࠾༻

͢Δɽ

j = m

i=1

Dt(i)δji (5.6)

δji=

1 if yi=hj(xi) 0 if yi=hj(xi)

j >0.5ͷ৔߹ɼֶशΛऴྃ͢ΔɽͦΕҎ֎ͷ৔߹͸ɼαtΛࣜ(5.7)Ͱܭࢉ͢Δɽ αt= 1

2ln1−t

t (5.7)

ͦͯ͠ɼֶशσʔλͷ֤αϯϓϧͷॏΈΛࣜ(5.8)ͷΑ͏ʹߋ৽͢Δɽ͜͜Ͱht͕ޡͬ

ͨαϯϓϧͷॏΈ͕ߴ͘ͳΔΑ͏ʹߋ৽͞ΕΔɽ

Dt+1(i) = Dt(i) exp(−αtyiht(xi))

Zt (5.8)

Zt ͸ਖ਼نԽ߲Ͱɼ

iDt+1(i) = 1Λຬͨ͢Α͏ʹௐ੔͞ΕΔɽࣜ(5.6)^͔Βࣜ(5.8)^· ͰͷܭࢉΛt= 1,· · ·, T^ʢ·ͨ͸j >0.5^{Λຬͨ͢௚લͷ}tʣ·Ͱߦ͍ɼ࠷ऴతʹࣜ(5.9) ʹࣔࣝ͢ผؔ਺Λੜ੒͢Δɽ

H(x) = sign _T

t=1

αtht(x)

(5.9)

5.4.2 ^ࣝผؔ਺

Adaboost^{๏ʹ͓͚Δࣝผؔ਺}hj ʹ͸ɼຊ࣮ݧͰ͸ܾఆ໦Λ༻͍ͨɽܾఆ໦ͱ͸ɼ؍

ଌσʔλΛڭࢣϥϕϧʹΑͬͯ໦ߏ଄ʹ෼ྨ͍ͯ͘͠ΞϧΰϦζϜͷҰछͰ͋Δɽܾఆ໦

ͷֶश͸ɼฏۉ৘ใྔʢΤϯτϩϐʔʣΛج४ʹࣝผྗʢ৘ใήΠϯʣ͕࠷େͱͳΔΑ͏

ͳ෼ྨϧʔϧʢଐੑͱ৚݅ʣΛٻΊͯࢬ෼͔Εͤ͞Δɽͦͯ͠ɼҰఆͷֶशճ਺·ͨ͸ܾ

ఆ໦શମͰҰఆͷࣝผྗ͕ಘΒΕͨஈ֊ͰֶशΛऴྃ͢Δɽຊ࣮ݧͰ͸ɼΤϯτϩϐʔͷ

ܭࢉʹAdaboost๏ʹ͓͚Δ֤αϯϓϧͷॏΈΛՃ͑ͯ෼ྨϧʔϧΛٻΊɼͦΕΒશͯΛ

1ͭͷऑࣝผثͱͯ͠༻͍ͨɽ͢ͳΘͪɼ2ճ໨Ҏ߱ͷֶशͰ͸ߋ৽͞Εͨαϯϓϧͷॏ

ΈΛ༻͍ܾͯఆ໦Λ࠶ߏஙͨ͠ɽ

5.4.3 ^{ଟΫϥε෼ྨ΁ͷԠ༻}

Adaboost^ɼSVM^ͱ΋ʹ2Ϋϥε෼ྨثͰ͋ΔͨΊɼ࡞ۀ಺༰ਪఆͷΑ͏ͳଟΫϥε෼

ྨ໰୊Ͱ͸ɼ1^{ରଞ෼ྨ๏΍}1^ର1^{෼ྨ๏Λ༻͍Δɽ}1ରଞ෼ྨ๏͸ɼ͋ΔΫϥεͱͦͷ ଞͷΫϥεશͯͷ2ΫϥεΛ෼ྨ͢ΔΑ͏ͳࣝผثΛશͯͷΫϥεͰߏங͠ɼଟΫϥε෼

ྨΛߦ͏ख๏Ͱ͋Δɽ1ର1෼ྨ๏͸ɼશͯͷΫϥε͔Βߟ͑ΒΕΔݶΓ2ΫϥεͷϖΞ Λ࡞ΓɼͦΕͧΕΛ෼ྨ͢ΔࣝผثΛߏங͠ɼ݁ՌΛ౷߹͢Δख๏Ͱ͋ΔɽຊݚڀͰ͸ɼ Adaboost^ʹ͸1^{ରଞ෼ྨ๏ɼ}SVM^ʹ͸1^ର1^{෼ྨ๏Λ༻͍ͨɽ}Adaboost^͸SVM^Α Γ΋ݸʑͷࣝผثͷֶशʹ͕͔͔࣌ؒΔ͕ɼऑࣝผثͷ૊Έ߹Θͤͱ͍͏ಛੑ͔Β1^ରଞ

෼ྨͰ͋ͬͯ΋ੑೳ͕ྼԽ͠ʹ͍ͨ͘Ίɼ1ରଞ෼ྨΛ࠾༻ͨ͠ɽ

5.4.4 ֶशͷྲྀΕͱ࣌ؒํ޲΁ͷ֦ு

ਤ5.8ʹ࡞ۀ಺༰ਪఆͷॲཧͷྲྀΕΛࣔ͢ɽҎޙɼ࡞ۀ಺༰ΛSO^ʢService Operation^ʣ ͱදه͢Δɽ·ͣɼཁૉσʔλΛඍখ۠ؒΔt͝ͱʹ෼ׂͯ͠ಛ௃நग़Λ͢Δɽ࣍ʹɼη άϝϯτ௕kΔt(1 ≤k ≤K)ɼγϑτ௕ΔtͰηάϝϯτ͝ͱʹ·ͱΊͨಛ௃ྔΛٻΊ Δɽi൪໨ͷඍখ۠ؒΛ։࢝ͱ͢Δɼηάϝϯτ௕kΔtʹର͢Δಛ௃ྔϕΫτϧΛf(i, k) ͱ͠ɼηάϝϯτ௕kΔt^ʹ͓͚Δf(i, k)^ͷू߹ΛTk ͱ͓͘ɽଓ͍ͯɼશͯͷSO_m^ͱ k^ʹରͯ͠TkͱSOͷਖ਼ղϥϕϧΛ༻͍ͯɼSO_m͔൱͔Λ൑ఆ͢ΔࣝผثCm,kΛલड़

ͷAdaboost๏ͱSVMʹΑΓֶश͢Δɽਪఆσʔλ΋ಉ༷ʹηάϝϯτԽ͠ɼ֤ࣝผث

Cm,k ͷ໬΋Β͠͞Λද͢είΞSi,Cm,k ΛٻΊΔɽͦͯ͠ɼҎԼͷࣜͰSOผʹॏΈ෇

͚૯࿨Li.mΛٻΊɼLi,mͷ໬΋ߴ͍SO_mΛਪఆ͞ΕͨSOͱ͢Δɽ

Li,m= K k=1

wm,kSi,Cm,k (5.10)

͜͜Ͱɼwm,k ͸֤SOͷ௕͞ͷ෼෍͔ΒٻΊΒΕΔॏΈͰ͋Δຊ࣮ݧͰ͸ɼΔt^Λ5

ਤ5.8 ࡞ۀ಺༰ਪఆͷΞϧΰϦζϜͷྲྀΕɽ

ඵɼK^Λ8ʹઃఆͨ͠ɽ

5.5 ࣮ݧ

ද5.1ʹࣔ͢8छྨͷ࡞ۀ಺༰ʢSOʣΛਪఆ͢Δ࣮ݧΛߦͬͨɽ࣮ݧ৚݅Λද5.7ʹ

ࣔ͢ɽֶशɾධՁʹ࢖༻ͨ͠σʔλ͸ܭ2019^αϯϓϧͰɼclosed^{৚݅Ͱ͸ֶशͱධՁʹ} ಉ͡σʔλΛ࢖༻͠ɼopen^৚݅Ͱ͸2019^{αϯϓϧΛ࣌ؒଳผʹ}10^{෼ׂͯ͠ަࠩݕఆʹ} ΑͬͯධՁͨ͠ɽಛ௃ྔ͸ද5.7ͷ4छྨͰɼ૊Έ߹ΘͤΛม͑ͯਪఆਫ਼౓Λൺֱͨ͠ɽ

ࣝผث͸ɼఏҊ͢ΔAdaboost๏ͱɼൺֱͱͯ͠SVMΛ༻͍ͯఏҊख๏ͷ༗ޮੑΛௐࠪ

ͨ͠ɽධՁई౓͸ɼ֤SOͷద߹཰ͱ࠶ݱ཰ͷௐ࿨ฏۉʢF஋ʣΛͦͷSOͷਪఆਫ਼౓ͱ

ͯ͠༻͍ͨʢࣜ(5.11)^ʙࣜ(5.13)^ʣɽ

SO_x^ͷద߹཰[%] = SO_xΛਖ਼͘͠ਪఆͨ͠αϯϓϧ਺

SO_x^{ͱਪఆͨ͠αϯϓϧ਺} (5.11) SO_xͷ࠶ݱ཰[%] = SO_xΛਖ਼͘͠ਪఆͨ͠αϯϓϧ਺

SO_xͷਅ஋ͷαϯϓϧ਺ (5.12) SO_x^{ͷਪఆਫ਼౓}[%] = 2∗SO_x^ͷద߹཰∗SO_x^ͷ࠶ݱ཰

SO_x^ͷద߹཰+ SO_x^ͷ࠶ݱ཰ (5.13)

ද5.7 ^࣮ݧ৚݅ɽ ਪఆ͢ΔSO ^ද5.1^ͷ8^छྨ

σʔλ਺ ֶशɾධՁσʔλܭ2019^αϯϓϧ

ඃݧऀ਺ 2໊

ಛ௃ྔ

(1)ηϯασʔλʹجͮ͘ಛ௃ྔʢ19࣍ݩʣ˞ද5.4 (2)ձܭσʔλʹجͮ͘ಛ௃ྔʢ12࣍ݩʣ˞ද5.6

(3)^{ྔࢠԽൃ࿩཰ʢ}8^࣍ݩʣ˞ද5.2 (4)ൃ࿩ͷ༗ແʹؔ͢Δಛ௃ྔʢ9^࣍ݩʣ˞ද5.2 ηάϝϯτ௕ 5^ɼ10^ɼ· · ·^ɼ40^ඵʢܭ8^छྨʣ

γϑτ௕ 5ඵ

ֶश৚݅ ࣌ؒଳผ10෼ׂަࠩݕఆ

5.5.1 ࣮ݧ݁Ռ

ਤ5.9ʹɼಛ௃ྔͱηάϝϯτ௕ͷ৚݅ΛมԽͤͨ͞৔߹ͷਪఆਫ਼౓ͷมԽΛࣔ͢ɽη ϯαɼձܭɼൃ࿩཰ɼ࿩ऀ͸ͦΕͧΕද5.7ͷಛ௃ྔͷ(1)ʙ(4)ʹ૬౰͢ΔɽԻ੠σʔλ ʹΑΔಛ௃ྔͷ༗ແʹରͯ͠ੑೳͷൺֱΛ͍ͯ͠Δɽԣ࣠͸ࣝผثΛ౷߹͢Δηάϝϯτ

௕ͷൣғͰ͋Δɽࠨ͔Βɼηάϝϯτ௕5ඵͷࣝผثͷΈͷ݁Ռɼಉ5ඵͱ10ඵͷ݁Ռ Λ౷߹ɼಉ5^ඵɼ10^ඵɼ15ඵͷ݁ՌΛ౷߹ͨ͠৔߹ͱଓ͖ɼҰ൪ӈ͕5^ඵ͔Β40^ඵ·

Ͱͷ8छྨͷηάϝϯτ௕ʹΑΔࣝผثͷ݁ՌΛ౷߹ͨ͠৔߹Ͱ͋ΔɽϞσϧͷֶश͸

Adaboost๏Ͱopen৚݅Ͱߦͬͨ΋ͷͰ͋Δɽಛ௃ྔʹ͍ͭͯ͸ɼηάϝϯτ௕ͷ৚݅

͕΋ͬͱ΋ྑ͍৔߹ɼηϯαͱձܭσʔλʹجͮ͘ಛ௃ྔͷΈͰ͸ਪఆਫ਼౓͕35.9%Ͱ

͋ΓɼྔࢠԽൃ࿩཰ʢਤதͷൃ࿩཰ʣΛՃ͑Δ͜ͱͰ45.9%ɼ͞Βʹൃ࿩ͷ༗ແʹؔ͢Δ ಛ௃ྔʢਤதͷ࿩ऀʣΛՃ͑Δ͜ͱͰ50.1%ʹվળͨ͠ɽ·ͨɼ͍ͣΕͷಛ௃ྔͷ૊Έ

߹Θͤʹ͍ͭͯ΋ɼ౷߹͢Δηάϝϯτ௕Λଟ͘͢Δ͜ͱʹΑͬͯਪఆਫ਼౓ͷվળ͕ݟΒ Εͨɽ

ਤ5.10ʹɼAdaboost๏ͷൺֱͱֶͯ͠शϞσϧΛSVMʹͨ͠৔߹ͷਪఆਫ਼౓ͷมԽ Λࣔ͢ɽಛ௃ྔ͸Adaboost๏Ͱ࠷΋݁Ռͷྑ͔ͬͨɼηϯαɼձܭɼྔࢠԽൃ࿩཰ɼൃ

࿩ͷ༗ແʹؔ͢Δಛ௃ྔͰ͋ΔɽఏҊख๏ΛSVMͰߦ͏৔߹͸ɼSVRʢSupport Vector

RegressionʣΛ༻͍ͯճؼϞσϧΛֶश͠ɼݸʑͷSOʹ͍ͭͯ༧ଌ֬཰ΛٻΊͯ݁Ռ

౷߹ͨ͠ɽ݁Ռ͸34.3%ͱͳΓɼਪఆਫ਼౓ɼηάϝϯτ௕ͷ݁Ռ౷߹ʹΑΔվળ෯΋

Adaboost๏ΑΓ௿͘ͳͬͨɽ

ද 5.8 ʹɼݸʑͷ࡞ۀ಺༰ͷਪఆ݁Ռʹ͍ͭͯ confusion matrix Λࣔ͢ɽ৚݅͸

Adaboost๏Ͱ࠷΋ྑ͔ͬͨ৔߹Ͱ͋Δɽελοϑͱձ࿩ʢSO No.2ʣɼ഑હʢSO No.5ʣɼ ย෇͚ʗηοςΟϯάʢSO No.7ʣ͸ൺֱతྑ͘ਪఆͰ͖͍ͯΔɽ͜ΕΒ͸ɼԻ੠σʔ λʹΑΔಛ௃ྔΛՃ͑Δ͜ͱͰ10ϙΠϯτҎ্վળ͞Ε͍ͯͨɽ͓٬͞Μͱձ࿩ʢSO No.1^{ʣɼҠಈʗӡൖʢ}SO No.8ʣ͸௿͍ɽશମతͳ܏޲ͱͯ͠ɼ઀٬࡞ۀʢSO No.1^ɼ3^ʙ 6ʣɼඇ઀٬࡞ۀʢSO No.2ɼ7ɼ8ʣಉ࢜Ͱ۠ผ͕͔ͭͣʹޡΔ܏޲͕ߴ͔ͬͨɽ઀٬࡞ۀ ʗඇ઀٬࡞ۀͱ͍͏2෼ྨͷ৔߹͸ਪఆਫ਼౓͕83%ͱͳΓɼ઀٬ʗඇ઀٬࡞ۀͷ෼ྨͰ

͋Ε͹͋Δఔ౓ग़དྷ͍ͯΔͱݴ͑Δɽ

ਤ5.9 ಛ௃ྔʹΑΔਪఆਫ਼౓ͷҧ͍ɽ

ද5.8 ^ਪఆ݁ՌͷConfusion Matrix^ɽ

ਅ஋ͷ ਪఆ͞Εͨ࡞ۀ಺༰ ద߹཰

࡞ۀ಺༰ 1 2 3 4 5 6 7 8 [%]

1 8 1 16 23 33 3 1 0 9.4

2 0 174 15 1 8 2 44 1 71.0

3 11 21 115 27 44 0 50 2 42.6

4 0 13 9 138 67 17 40 0 48.6

5 2 0 40 30 250 17 21 2 69.1

6 0 5 25 23 6 33 9 0 32.7

7 0 34 51 17 86 3 354 17 63.0

8 0 2 8 4 5 2 70 19 17.3

࠶ݱ཰[%] 38.1 69.6 41.2 52.5 50.0 42.9 60.1 50.1 83.1

!"#

ਤ5.10 ࣝผثʹΑΔਪఆਫ਼౓ͷҧ͍ɽ

5.5.2 ߟ࡯ͱ՝୊

ྔࢠԽൃ࿩཰͸શମతʹੑೳΛվળ͓ͯ͠Γɼ࡞ۀ಺༰ਪఆͷಛ௃ྔͱͯ͠༗ޮͰ͋

Δͱݴ͑ΔɽAdaboost๏ʹ͓͍ͯϞσϧͷֶशʹ༻͍ܾͨఆ໦ͷ෼ྨϧʔϧʹண໨͢Δ ͱɼଟ͘ͷ࡞ۀ಺༰ʹ͍ͭͯ2ʙ5ஈ֊໨ͷ͖͍͠஋ͰٻΊͨൃ࿩཰͕෼ྨʹྑ͘ߩݙ͠

͍ͯͨɽൃ࿩ͷ༗ແʹؔ͢Δಛ௃ྔɼ͢ͳΘͪ࿩ऀΫϥεͷ৘ใΛՃ͑ͨ৔߹ɼελοϑ

ͱձ࿩ʢSO No.2^{ʣ΍Ҡಈɾӡൖʢ}SO No.8ʣͷਪఆਫ਼౓͕޲্͍ͯͨ͠ɽ͜Ε͸ɼ࿩ऀ

͕ݶఆ͞ΕΔΑ͏ͳ࡞ۀͰ͋ͬͨͨΊͰ͋Δɽʮελοϑͱձ࿩ʯʹ͍ͭͯ͸ଞͷैۀһ ͱͷൃ࿩͕ଟ͘ɼސ٬ͱͷൃ࿩͕গͳ͍ͱ͍ͬͨ෼ྨϧʔϧ͕Α͘ݟΒΕͨɽʮҠಈɾӡ ൖʯʹ͍ͭͯ͸ɼશͯͷൃ࿩͕গͳ͍ɼ͋Δ͍͸લޙʹϚΠΫΛ૷ணͨ͠ैۀһͷൃ࿩͕

͋Δɼͱ͍͏෼ྨϧʔϧ͕Α͘ݟΒΕͨɽͦͷΑ͏ͳ࡞ۀΛਪఆ͢Δ৔߹ʹ͸࿩ऀΫϥε ͷ৘ใ͸༗ޮੑ͕͋Δͱݴ͑Δɽ

ෳ਺ͷηάϝϯτ௕ͷ݁ՌΛ౷߹͢Δ͜ͱ͸Adaboost๏Ͱ͸༗ޮͰ͕͋ͬͨɼSVM Ͱ͸͋·Γػೳ͠ͳ͔ͬͨɽͦͷཧ༝ͱͯ͠ɼ࡞ۀผͷֶशσʔλ਺ʹ͹Β͖͕ͭେ͖

͘ɼಛʹ௕͍ηάϝϯτ௕ʹֶ͓͍ͯशσʔλ͕ۃ୺ʹগͳ͘ͳΓֶशͰ͖ͳ͍έʔε

͕͋ͬͨͨΊͱߟ͑ΒΕΔɽAdaboost๏Ͱ͸ֶशσʔλ਺ͷӨڹΛड͚ʹ͘͘ɼఏҊ ख๏΍࡞ۀ಺༰ਪఆʹ߹கͨࣝ͠ผثͱݴ͑Δɽ·ͨɼSVM ͸શͯͷಛ௃ྔΛֶशʹ

࢖༻͢ΔͨΊɼಛ௃ྔ͝ͱͷ෼ྨੑೳ΋࡞ۀ಺༰ਪఆͷਫ਼౓ʹӨڹ͢Δͱߟ͑ΒΕΔɽ

Adaboost๏Ͱ͸ɼ෼ྨੑೳͷߴ͍ಛ௃ྔʹॏ఺Λஔͨ͘ΊɼຊσʔλͷΑ͏ʹಛ௃ྔͷ

ਫ਼౓͕௿͘ͳΔ͜ͱ͕༧૝͞ΕΔ৔߹ʹ͸ɼAdaboost๏ͷํ͕ྑ͍ͱߟ͑ΒΕΔɽ close৚݅Ͱ͸Adaboost๏ʹͯਪఆਫ਼౓͕98%Ͱ͋Γɼopen৚݅Ͱ΋ϥϯμϜαϯ ϓϦϯά͢Δ৔߹͸ਪఆਫ਼౓͕࠷େͰ88%ʹͳΔ͕ɼຊ࣮ݧͷΑ͏ʹσʔλΛ࣌ؒଳʹ Αͬͯ෼ׂͨ͠৔߹͸50%·ͰམͪࠐΜͩɽ͜Ε͸ɼಉ͡࡞ۀͰ΋࣌ؒଳʹΑͬͯ෼෍

ʹେ͖ͳภΓ͕͋Δͱߟ͑ΒΕΔɽ͢ͳΘͪɼ࡞ۀ಺༰ਪఆΛݱ৔ʹద༻͢Δ৔߹ɼैۀ һ΍࡞ۀ೔࣌ʹΑͬͯੑೳʹ͹Β͖͕ͭग़ͯ͠·͏ڪΕ͕͋Δɽ͞Βʹਪఆਫ਼౓ΛੑೳΛ

޲্ͤ͞Δʹ͸ɼಛ௃ྔͷ෼෍ͷภΓΛͳ͘͢Α͏ʹֶशσʔλΛ૿΍͠ɼຬวͳ͘ਪఆ Ͱ͖ΔϞσϧ͕ඞཁͰ͋Δͱݴ͑Δɽ

ୈ 6 ^ষ

݁࿦

6.1 ੒Ռ

ຊݚڀͰ͸ɼैۀһͷۀ຿தͷԻ੠σʔλʹର͢Δൃ࿩۠ؒݕग़ɼ࿩ऀΫϥε෼ྨͱɼ

͜ΕΒͷ݁ՌΛ༻͍ͨ࡞ۀ಺༰ਪఆʹ͍ͭͯఏҊͨ͠ɽϨετϥϯʹ͓͚Δैۀһͷۀ຿

தͷԻ੠σʔλ͸ɼଟ͘ͷࡶԻ͕ॏ৞͓ͯ͠ΓɼҠಈ΍࡞ۀʹΑͬͯ࿥Ի؀ڥ͕มԽͨ͠

Γɼ࡞ۀʹΑͬͯൃ࿩ύϫʔ͕ҟͳΔͳͲɼॲཧͷ೉͍͠σʔλͰ͋ͬͨɽ͜Εʹର͠ɼ

ൃ࿩۠ؒݕग़΍࿩ऀΫϥε෼ྨΛߦ͍ɼطଘख๏ͷ૊Έ߹ΘͤʹΑͬͯɼ࿩ऀ΍؀ڥͷҧ

͍ʹΑΒͣɼҰఆͷੑೳ͕ಘΒΕΔ͜ͱΛ֬ೝͨ͠ɽ࿩ऀΫϥε෼ྨʹ͍ͭͯ͸ɼෆಛఆ

࿩ऀͷΫϥεΛѻ͏ͱ͍͏఺Ͱ࿩ऀࣝผΑΓ΋೉͍͠λεΫͰ͕͋ͬͨɼಛఆ࿩ऀʹద͢

Δ෼ྨثͱෆಛఆ࿩ऀʹద͢Δ෼ྨثΛ໌Β͔ʹͨ͠ɽ྆ऀͷ૊Έ߹ΘͤʹΑͬͯɼ͞Β ʹੑೳΛվળ͢Δ͜ͱ͕ظ଴͞ΕΔɽ࡞ۀ಺༰ਪఆʹ͍ͭͯ͸ɼൺֱత࣌ؒͷ௕͍࡞ۀΛ

ࣝผ͢ΔͨΊɼಛ௃ྔ΍ࣝผثʹ࣌ؒత֓೦ΛऔΓೖΕͨɽ·ͨɼൃ࿩Λͱ΋ͳ͏࡞ۀ͕

ଟ͘ɼԻ੠σʔλ͔Βൃ࿩۠ؒͱ࿩ऀΫϥε৘ใΛಛ௃ྔͱͯ͠Ճ͑ͯਪఆΛߦͬͨɽͦ

ΕͧΕͷಛ௃ྔʹҰఆͷਫ਼౓͕͋Γɼ͔ͭ࡞ۀʹݟΒΕΔൃ࿩ͷׂ߹͕ଟ͚Ε͹ɼ࡞ۀ಺

༰ਪఆʹ༗ޮͰ͋Δ͜ͱΛࣔͨ͠ɽ

αʔϏε޻ֶʹ͓͍ͯɼैۀһͷߦಈ෼ੳʹԻ੠Λ༻͍ͨࣄྫ͸গͳ͍ɽಛʹɼԻ੠࿥

Իʹର͢ΔϓϥΠόγʔͷ໰୊΍ɼԻ੠σʔλͷղऍͷ೉͠͞ͳͲɼಋೖ͢ΔͨΊͷোน

͸௿͘ͳ͍ɽຊݚڀ͸ͦͷதͰɼΩʔϫʔυ΍ൃ࿩಺༰ͳͲԻ੠ͷ۩ମతͳ৘ใʹ͸৮Ε

ͣɼൃ࿩۠ؒ΍࿩ऀͷ৘ใͷΈΛ༻͍Δ͜ͱͰɼ͜ΕΒͷ໰୊΁ͷղܾΛࢼΈͨɽൃ࿩ͷ

ׂ߹΍࿩ऀͷ৘ใͱߦಈʹؔ܎ੑ͕ݟ͍ग़ͤΔͳΒ͹ɼϨετϥϯ؀ڥʹґଘͤͣɼଟ͘

ͷ࿑ಇ؀ڥͰԻ੠৘ใΛ׆༻͢Δ͜ͱ͕Ͱ͖Δͱߟ͑ΒΕΔɽಛʹɼैۀһɾސ٬ͷ૒ํ

͕࿩ऀͱͯ͠ଘࡏ͢Δ৔߹ʹ͸ɼຊख๏͕ͦͷ··ద༻Ͱ͖Δɽձܭσʔλ΋࡞ۀ಺༰ਪ ఆͷڧྗͳཁૉͰ͋Δ͕ɼͲͷݱ৔Ͱ΋ಉ༷ͷձܭσʔλΛಘΒΕΔ༁Ͱ͸ͳ͍ɽҰํɼ Ի੠σʔλ΋ैۀһͷൃ࿩཰͕ߴ͍͜ͱ͕ٻΊΒΕΔͨΊɼ૒ํͷෆ଍఺Λิؒ͢Δ͜ͱ ͰΑΓଟ͘ͷݱ৔ʹ࡞ۀ಺༰ਪఆΛద༻͢Δ͜ͱ͕๬·ΕΔɽ

6.2 ՝୊ͱల։

ຊݚڀͰఏҊͨ͠ݸʑͷٕज़͸ɼݱ৔΁ͷϑΟʔυόοΫΛ૝ఆͨ͠৔߹ʹ͸ɼ͍ͣΕ

΋ੑೳͷ޲্͕ٻΊΒΕΔɽൃ࿩۠ؒݕग़ʹ͍ͭͯ͸ɼϚΠΫϩϗϯΛ૷ணͨ͠ैۀһʹ

͍ͭͯ͸ྑ͘நग़Ͱ͖͍͕ͯͨɼͦΕҎ֎ͷൃ࿩ʹ͍ͭͯ͸े෼ʹݕग़Ͱ͖͍ͯͳ͍ɽ͜

ΕΒ͸ɼඇൃ࿩۠ؒ΋ަ͑ͯҰ୴நग़͔ͯ͠Βਫ਼ࠪ͢Δͱ͍͏Α͏ͳɼҟͳΔΞϓϩʔν

͕ٻΊΒΕΔɽ͜Ε͸ɼ࿩ऀΫϥε෼ྨͷΞϓϩʔνʹ΋େ͖ؔ͘ΘΔɽ࿩ऀΫϥε෼ྨ

ͷεςοϓͰ͸ɼ୯ʹൃ࿩۠ؒΛ෼ྨ͢ΔͷͰ͸ͳ͘ɼ৽ͨʹඇൃ࿩΍ࡶԻ۠ؒͷϞσϧ Λ࡞ΔͳͲͯ͠ΑΓϛΫϩʹൃ࿩Λ෼ੳ͍ͯ͘͠ํ͕ࣜ๬·ΕΔɽݸʑͷਪఆٕज़͸ਪఆ ϞσϧΛඞཁͱ͢Δ͕ɼϞσϧΛֶश͢ΔͨΊͷڭࢣσʔλ͸ϓϥΠόγʔ΍σʔλ؅ཧ ͷ໰୊͔Β࠷௿ݶʹ͢Δඞཁ͕͋ΔɽΑΓগྔͷڭࢣσʔλͰֶशͰ͖Δɼ͋Δ͍͸దԠ తͳϞσϧֶशख๏͕ٻΊΒΕΔɽ·ͨɼݸʑͷਪఆٕज़ͷग़ྗ͸ɼݱঢ়ͷΑ͏ͳܾఆ࿦

తͳग़ྗΑΓ΋֬཰తͳग़ྗͷํ͕ɼΑΓԠ༻ൣғ͕޿͘ͳΔͱߟ͑ΒΕΔɽ

࡞ۀ಺༰ਪఆʹ͍ͭͯ΋े෼ͳਫ਼౓͕ಘΒΕ͍ͯΔͱ͸ݴ͑ͳ͍͕ɼಛ௃ྔͷภΓΛݮ Β͢ͳͲֶशσʔλΛద੾ʹઃఆ͢Ε͹ՄࢹԽʹे෼ͳੑೳ͕ಘΒΕΔ͜ͱ΋ࣔ͞Εͨͨ

ΊɼσʔλΛ૿΍͢͜ͱͰ൚༻ੑΛߴΊΔ͜ͱ͕࣍ͷεςοϓͱͳΔɽಛ௃ྔ΍ࣝผثʹ

͍ͭͯ΋ݕ౼ͷ༨஍͸͋Δ͕ɼ৽ͨͳΞϓϩʔνͱͯ͠ɼސ٬ߦಈϞσϧͷݕ౼͕ڍ͛Β ΕΔɽސ٬ߦಈϞσϧͱ͸ɼސ٬ࢹ఺ͰैۀһͷߦಈΛਪఆ͢ΔϞσϧͰ͋Δɽձܭσʔ λͷސ٬਺΍ೖళɾୀళͷλΠϛϯάʹΑͬͯ͋Δఔ౓࡞ۀ಺༰ͷ෼෍͕༧ଌͰ͖ɼސ٬

ߦಈϞσϧͷߏஙʹͭͳ͕ΔɽຊݚڀͰఏҊͨ͠ैۀһͷߦಈϨϕϧͰͷϞσϧͱސ٬ߦ ಈϞσϧͱ༥߹͢Δ͜ͱʹΑͬͯɼ࡞ۀ಺༰ਪఆͷੑೳ͕ߴΊΒΕΔͱߟ͑ΒΕΔɽ

ຊݚڀͰఏҊͨ͠࿮૊ΈΛݱ৔Ͱద༻͢ΔͨΊʹ͸ɼ࡞ۀ಺༰ਪఆͷ݁ՌΛͲͷΑ͏ʹ ݱ৔ʹఏࣔ͢Δ͔ɼͦͯ͠ɼͲͷఔ౓ͷՁ஋͕͋Δ͔ݕূ͢Δඞཁ͕͋Δɽͦͷ্Ͱɼਪ ఆͷ಺༰΍ඞཁͱ͢Δॲཧ࣌ؒɼਫ਼౓ʹ͍ͭͯվΊͯٞ࿦͢Δ͜ͱ͕๬·͍͠ɽଞళฮ΍

ۀछ΁࿮૊ΈΛల։͢Δ͜ͱ͸໪࿦Ͱ͋Δ͕ɼ࡞ۀ಺༰ਪఆʹ͍ͭͯ͸ैۀһͷۀ຿Ҏ֎

ͷߦಈʹ͍ͭͯ΋ద༻͢Δ͜ͱ͕ظ଴͞ΕΔɽ

ँࣙ

ຊݚڀΛߦ͏ʹ͋ͨΓɼݚڀํ਑ͷ૬ஊɼऴ࢝ஆ͔͍ޚࢦಋΛ௖͖·ͨ͠ذෞେֶ޻ֶ

෦ Ԡ༻৘ใֶՊɼ଎ਫޛڭतɼాଜ఩࢚ॿڭʹਂ͘ײँ͍ͨ͠·͢ɽ·ͨɼຊ࿦จʹ͍ͭ

ͯɼ༗ҙٛͳޚࢦఠɼޚॿݴΛࣀΓ·ͨ͠ɼذෞେֶ޻ֶ෦ɹిؾిࢠɾ৘ใ޻ֶՊɹ৘

ใίʔεɼҏ౻তڭतɼԣా߁੒ڭतɼח෦ߒڭतʹਂ͘ײँ͍ͨ͠·͢ɽ

ຊݚڀͷ͖͔͚ͬΛ༩͑ͯͩͬͨ͘͞࢈ۀٕज़૯߹ݚڀॴɼαʔϏε޻ֶݚڀηϯλʔ ͷଂా෢ࢤࢯɼఱ໨ོฏࢯɼՃ౻ङເࢯʹ͸ɼݚमޙ΋ऴ࢝ଟେͳޚࢦಋɼޚॿݴΛ͍ͨ

͖ͩɼਂ͘ײँக͠·͢ɽ·ͨɼڞಉͰݚڀΛߦΘͤͯ௖͖·ͨ͠࢈ۀٕज़૯߹ݚڀॴɼ αʔϏε޻ֶݚڀηϯλʔͷօ༷ʹ͸େม͓ੈ࿩ʹͳΓ·ͨ͠ɽਂ͘ײँக͠·͢ɽ

·ͨɼݚڀ༻ͷԻ੠σʔλΛఏڙͯ͠௖͍͕ͨΜ͜ϑʔυαʔϏεגࣜձ༷ࣾɼגࣜձ

ࣾγςΟʔɾΤεςʔτ༷ɼࠎಋԻσʔλͷܭଌʹ͋ͨΓࠎ఻ಋϚΠΫϩϗϯͷ։ൃɾఏ ڙΛͯ͠௖͍ͨגࣜձࣾςϜίδϟύϯ༷ɼԻڹϞσϧ༻ͷԻ੠σʔλΛఏڙͯ͠௖͍ͨ

ϝσΟΞυϥΠϒגࣜձ༷ࣾɼ࣌ࠁಉظγεςϜͷ։ൃɾఏڙΛͯ͠௖͍ͨϚϧςΟεʔ ϓגࣜձ༷ࣾʹ͸ɼް͘ޚྱਃ্͛͠·͢ɽ

ͦͯ͠ɼ଎ਫɾాଜݚڀࣨͷαʔϏε෼໺ͷݚڀ൝ͷϝϯόʔͰ͋ΔɼӉ໺ଠٱ࠸ࢯɼ

઒੉ప໵ࢯɼ໺৲߂໵ࢯɼՃౡ୎ຏࢯɼ઒ౢେٛࢯɼٶ઒Ճಸࢠࢯɼࢁాࢻ৫ࢯɼ୮Ӌߴ

େࢯʹ͸ɼθϛ΍ϛʔςΟϯάͳͲͰوॏͳޚҙݟɼޚॿݴΛ௖͖ɼ͔ͭಉ͡ݚڀ෼໺ͱ

ͯ͠ଟ෼ʹޚॿྗ͍͖ͨͩ·ͨ͠ɽਂ͘ײँக͠·͢ɽ

࠷ޙʹɼ଎ਫɾాଜݚڀࣨͷݱ໾ੜͷօ༷ͱɼOB^ɾOGͷํʑʹ͸ɼ௕೥ͷݚڀੜ׆

΍θϛΛ௨ͯ͠وॏͳޚҙݟɼޚॿݴΛ௖͖·ͨ͠ɽҎ্օ༷ʹਂ͘ײँக͠·͢ɽ

ࢀߟจݙ

[1] ੺দװ೭ɼ৽Ҫຽ෉ɼ಺౻ߞɼଜ্ً߁ɼ٢ຊҰึɼαʔϏε޻ֶ ʕ51^ͷٕज़ͱ࣮

ફʕ,ே૔ॻళ, 2012ɽ

[2] T. Tomiyama, ”A manufacturing paradigm toward the 21st century,” Computer Aided Engineering, Vol.4, pp.159-178, 1997.

[3] R. Tenmoku, R. Ueoka, K. Makita, T. Shinmura, M. Takehara, S. Tamura, S.

Hayamizu, T. Kurata: ”Service-operation estimation in a Japanese restaurant using multi-sensor and POS data,” Proc. Advances in Production Management Systems 2011 Conference, Parallel 3-4: 1, Stavanger, Norway, Sep. 2011.

[4] B. Hartman, C. Schauer, and N. Link, ”Worker behavior interpretation for flexi-ble production,” Proc. World Academy of Science, Engineering and Technology, Vol.3, pp.494-502, Roma, Italy, Apr. 2009.

[5] S. Chae and T. Yoshida, ”A study of safety management using working area information on construction site,” Proc. the 25th International Symposium on Automation and Robotics in Construction, pp.292-299, Lithuania, June. 2008.

[6] അ৔७ࢠɼʠհޢอݥ੍౓վਖ਼ʹΑΔέΞϚωδϝϯτۀ຿ྔͷมԽʡɼઐमਓؒՊֶ

࿦ूɼVol.2^ɼno.2^ɼpp.99-111^ɼMar. 2012^ɽ

[7] άϧʔϓϗʔϜɾ࿝݈ࢪઃͰͷλΠϜελσΟௐࠪɼhttp://www.jpha.or.jp/

sub/pdf/menu04_5_09_03.pdf^ɽʢ2016-1-6ࢀরʣ

[8] ౦໺ఆ཯ɼ౵Ҫ޹ࢠɼʠපӃซઃܕೕࣇӃͷೖॴೕ༮ࣇͷঢ়ଶ૾ͱఏڙ͞ΕͨέΞ࣮

ଶʹؔ͢ΔݚڀʡɼܦӦͱ৘ใɿ੩ԬݝཱେֶɾܦӦ৘ใֶ෦/ֶใɼVol.23ɼno.2ɼ pp.1-12^ɼMar. 2011^ɽ

[9] ࢁాݑ࢚ɼബ༤ేɼ෢ాਅقɼԬాࢤຑɼਗ਼ਫࠤ஌ࢠɼੴҪ๛ܙɼத઒ཬܙɼদଜହ ࢤɼ໺ా༞ࢠɼ௕૔ढ़໌ɼੴݪݠɼେ໺Ώ͏ࢠɼʠ؃ޢۀ຿ΛଌΔ–^{ແਓλΠϜελ} σΟʡɼੜମҩ޻ֶɼVol.48ɼNo.6ɼpp.513-516ɼDec. 2010ɽ

[10] F. Naya, R. Ohmura, F. Takayanagi, H. Noma and K. Kogure, K, ”Workers’

routine activity recognition using body movements and location information,”

Proc of the 10th International Symposium on Wearable Computers, pp.105-108, Montreux, Oct. 2006.

[11] R. Ohmura, F. Naya, H. Noma, N. Kuwahara, T. Toriyama, and K. Kogure,

”Practical design of a sensor network for understanding nursing activities,” Proc.

the 31st IEEE Conference on Local Computer Networks, pp.615-622, Tampa, FL, Nov. 2006.

[12] ^{ࣇۄঢฏɼଜ઒ࡾ࿠ɼ}ʠϑΝϛϦʔϨετϥϯਥ๪ʹ͓͚Δௐཧਓͷಈ࡞ɾಈઢʹؔ͢

Δௐࠪݚڀ(^ݐஙܭը)ʡɼ೔ຊݐஙֶձதࠃࢧ෦ݚڀใࠂूɼ೔ຊݐஙֶձ, Vol.27^ɼ pp.589-592^ɼMar. 2004.

[13] দ೾੖ਓɼʠߦಈ؍࡯ʹΑΔαʔϏεݱ৔վળʢҿ৯ళɼమಓɼϗςϧʣʡɼ೔ຊਓؒ

޻ֶձୈ51ճେձɼJune. 2010ɽ

[14] ஛தؽɼ৽ଜ໠ɼʠ֎৯࢈ۀʹ͓͚ΔαʔϏε޻ֶͷ࣮ફʡɼୈ51ճਓ޻஌ೳֶձେ

ձɼ2010^ɽ

[15] ߕ ൹ Έ ͷ Γ ɼʠ Ի ੠ ෼ ੳ ʹ Α Δ Ի ੠ ෼ ੳ Ԡ ର ඼ ࣭ ͷ ޮ ཰ త ޲ ্ ʡɼhttp://

www.solco.co.jp/File/d0000/index/2Jirei_Onseibunseki.pdf^ɼNTT ι ϧ ίɼ2010ɽʢ2016-1-6ࢀরʣ

[16] ࠤ໺ਈ໵ɼനҪ߁೭ɼ౦໺߃೭ɼೋළਖ਼ɼʠԻ੠ղੳʹ΋ͱͮ͘ςΩετ෼ੳͷՄೳ

ੑͱࠓޙͷల๬ʡɼࡾඛ૯߹ݚڀॴʮॴใʯɼNo.52^ɼpp.74-88^ɼ2010^ɽ

[17] M. Abhishek Pandharipande and S.K. Kopparapu, ”A novel approach to identify problematic call center conversations,” Proc. the 9th International Joint Confer-ence on Computer SciConfer-ence and Software Engineering, pp.1-5, Bangkok, Thailand, June. 2012.

[18] K. Ishiguro, T. Yamada, S. Araki and T. Nakatani, ”A probabilistic speaker clus-tering for DOA-baseddiarization,” Proc. 2009 IEEE Workshop on Applications of Signal Processing to Audio and Acoustics, pp.241-244, New Paltz, NY, Oct.

2009.

[19]ʰQCαʔΫϧ׆ಈӡӦͷجຊʦ৽൛ʧʱQCαʔΫϧຊ෦ɼ೔ຊՊֶٕज़࿈ໍɼJul.

1997.

[20] ౻໺ਅɼʠ੡଄ۀʹ͓͚Δখूஂ׆ಈͷ໾ׂͱػೳʡɼ࿑຿ཧ࿦ֶձࢽɼNo.14^ɼ pp.177-190^ɼMar. 2005^ɽ

[21] F. Tomohiro, R. Tenmoku, T. Okuma, R. Ueoka, M. Takehara and T. Ku-rataɼ”Improving service processes based on visualization of human-behavior and POS data: a Case Study in a Japanese Restaurant,” Proc. the 1st International Conference on Serviceology, Tokyo, Japan, Oct. 2013.

[22] Ϙ Π ε σ ϡ ʔ α ʔ Π Ϡ ʔ Ϛ Π Ϋ ϩ ϗ ϯ ɼhttp://temco-j.cn/japanese/

products/audio/em_p.html^ɼTEMCO JAPAN^ɽʢ2016-1-6^ࢀরʣ

[23] M. Takehara, S. Tamura, S. Hayamizu, R. Tenmoku, T. Okuma, T. Fukuhara and T. Kurata: ”Measurement and analysis of speech data toward improving service in restaurant,” Proc. International Conference Oriental COCOSDA 2013,

pp.1-4, Gurgaon, India, Nov. 2013.

[24] ࠎ఻ಋϚΠΫࣜϊΠζΩϟϯηϦϯάγεςϜ౥ࡌɺߴԻ࣭Bluetoothϔουηοτɼ http://trinity.jp/products/jawbone/iconhd/, JAWBONE.ʢ2016-1-6ࢀরʣ [25] P. HeracleousɼJ. EvenɼC. IshiɼH. Ishigro and N. HagitaɼʠUsing alternate

acoustic sensors for speech communication and robust speech recognition^ʡɼ೔ຊ Իڹֶձ

[26] ౻ຊխਗ਼ɼʠԻ੠۠ؒݕग़ͷجૅͱ࠷ۙͷݚڀಈ޲ʡɼిࢠ৘ใ௨৴ֶձٕज़ݚڀใ ࠂɼVol.110, no.81, pp.7-12, June 2010ɽ

[27] Z. Song, T. Zhang, D. Zhang and T. Song, ”Voice activity detection using higher-order statistics in the teager energy domain,” Proc. International Conference on Wireless Communications and Signal Processing 2009, pp.1-5, Nanjing, Nov.

2009.

[28] ETSI ES 202 050 v.1.1.4, ”Speech processing, Transmission and Quality aspects (STQ), Advanced Distributed Speech Recognition (ADSR); Front-end feature extraction algorithm; Compression algorithms,” Nov. 2006.

[29] K. Ishizuka and T. Nakatani, ”Study of noise robust voice activity detection based on periodic component to aperiodic component ratio,” Proc. ISCA ITRW SAPA, pp.65-70, Pittsburg, USA, Sep. 2006.

[30] H. Wang, Y. Xu and M. Li, ”Study on the MFCC similarity-based voice activ-ity detection algorithm,” Proc. Artificial Intelligence, Management Science and Electronic Commerce 2011, pp.4391-4394, Deng Leng, China, Aug. 2011.

[31] T. Fukuda, O. Ichikawa and M. Nishimura, ”Long-term spectro-temporal and static harmonic features for voice activity detection,” Proc. Selected Topics in Signal Processing, IEEE, Vol.4, pp.834-844, Oct. 2010.

[32] େ௬యߦɼٶӬتҰɼԻ੠ࡶԻআڈʹؔ͢Δݚڀ ʵϥϯχϯάεϖΫτϧϑΟϧλ ʢRSFʣͷޮՌʵ

[33] G. Tzanetakis and P. Cook, ”Musical genre classification of audio signals,” IEEE Transactions on Speech and Audio Processing, Vol. 10, No. 5, pp. 293-302 (2002).

[34] T. Matsui and T. Tanabe, Comparative study of speaker identi

cation methods: dPLRM, SVM and GMM. IEICE transactions on information and systems, 89(3), pp. 1066-1073, 2006.

[35] D. A. Reynolds and R. C. Rose, Robust text-independent speaker identi

cation using Gaussian mixture speaker models. Speech and Audio Processing, IEEE Transactions on, 3(1), 72-83, 1995.

[36] S. H. Chen and Y. R. Luo, Speaker veri

cation using MFCC and support vector machine. In Proceedings of the Interna-tional MulitiConference of Engineers and Computer Scientists, Vol. 1, pp. 18-20,

2009.

[37] T. Yamada, L. Wang, and A. Kai, Improvement of distant-talking speaker identi cation using bottleneck features of DNN, In INTERSPEECH (pp. 3661-3664), 2013.

[38] ਫ໺׮೭ɼ஛಺৳Ұɼాଜ఩࢚ɼ଎ਫޛɼʡձٞ࿥ࣗಈ࡞੒γεςϜʹ޲͚ͨ࿩ऀࣝ

ผٕज़ͷݕ౼ʡɼୈ8ճ৘ใՊֶٕज़ϑΥʔϥϜ(FIT2009)^ɼE-047^ɼ2009^ɽ [39] Y. Lei, L. Burget and N. Scheffer, A noise robust i-vector extractor using vector

taylor series for speaker recognition. In ICASSP (pp. 6788-6791), 2013.

[40] P. Vincent, H. Larochelle, Y. Bengio and P. A. Manzagol, Extracting and com-posing robust features with denoising autoencoders, In Proceedings of the 25th international conference on Machine learning (pp. 1096-1103). ACM, 2008.

[41] P. Vincent, H. Larochelle, I. Lajoie, Y. Bengio and P. A. Manzagol, Stacked denoising autoencoders: Learning useful representations in a deep network with a local denoising criterion, The Journal of Machine Learning Research, 11, 3371-3408, 2010.

[42] Y. Freund and R. E. Schapire, ”A Decision-theoretic Generalization of On-line Learning and an Application to Boosting,” Jour. of Computer and System Sci-ences, Vol.55, No.1, pp.119-139, Aug. 1997.

ݚڀۀ੷

ֶज़ݚڀ࿦จ

1. ۀ຿Ի੠ͷൃ࿩۠ؒݕग़ʹΑΔ࡞ۀਪఆͷվળɽ

ɹ ஛ݪਖ਼ۣɼՃ౻ङເɼాଜ఩࢚ɼఱ໨ོฏɼଂా෢ࢤɼ଎ਫޛɽ ɹ ిࢠ৘ใ௨৴ֶձ࿦จࢽɼVol.J97-D,No.10,pp.1563-1571, Oct. 2014.

ࠃࡍձٞ

1. Service-operation estimation in a Japanese restaurant using multi-sensor and POS data.

ɹ Ryuhei Tenmoku, Ryoko Ueoka, Koji Makita, Takeshi Shinmura, Masanori Takehara, Satoshi Tamura, Satoru Hayamizu and Takeshi Kurata.

ɹ Advances in Production Management Systems International Conference 2011 (APMS2011), Jun. 2011.

2. The role of speech technology in service-operation estimation.

ɹ Masanori Takehara, Satoshi Tamura, Ryuhei Tenmoku, Takeshi Kurata and Satoru Hayamizu.

The 15th Oriental International Committee for the Co-ordination and Standardization of Speech Databases and Assessment Techniques (O-COCOSDA2011), Oct. 2011.

3. Toward improvement of SDR accuracy using LDA and query expansion for SpokenDoc.

ɹ Kiichi Hasegawa, Hideki Sekiya, Masanori Takehara, Taro Niinomi, Satoshi Tamura and Satoru Hayamizu.

ɹ The 9th NTCIR Workshop Meeting, pp.261-263, Dec. 2011.

4. Toward polyphonic musical instrument identification using example-based sparse representation.

ɹ Mari Okamura, Masanori Takehara, Satoshi Tamura, Satoru Hayamizu.

ɹ Asia-Pacific Signal and Information Processing Association Annual Sum-mit and Conference 2012 (APSIPA2012), Dec. 2012.

5. Statistical Voice Conversion using GA-based Informative Feature.

ɹ Kohei Sawada, Yoji Tagami, Satoshi Tamura, Masanori Takehara and Satoru Hayamizu.

ɹ Asia-Pacific Signal and Information Processing Association Annual Sum-mit and Conference 2012 (APSIPA2012), Dec. 2012.

6. Measuring and evaluating real service operations with human-behavior sensing:

a case study in a Japanese cuisine restaurant.

ɹ Tomohiro Fukuhara, Ryuhei Tenmoku, Takashi Okuma, Masanori Take-hara, and Takeshi Kurata.

ɹ The 19th Korea-Japan Workshop on Frontiers of Computer Vision (FCV2013), pp.113-116, 2013.

7. Spoken document retrieval using extended query model and web documents.

ɹ Kiichi Hasegawa, Masanori Takehara, Satoshi Tamura and Satoru Hayamizu.

ɹ The 10th NTCIR Conference, pp.608-611, Jun. 2013.

8. Improving service processes based on visualization of human-behavior and POS data: A case study in a Japanese restaurant.

ɹ Tomohiro Fukuhara, Ryuhei Tenmoku, Takashi Okuma, Ryoko Ueoka, Masanori Takehara, and Takeshi Kurata.

ɹ The 1st International Conference on Serviceology (ICServ2013), pp.1-8, Sep. 2013.

9. Measurement and analysis of speech data toward improving service in restau-rant.

ɹ Masanori Takehara, Satoshi Tamura, Satoru Hayamizu, Ryuhei Tenmoku, Takashi Okuma, Tomohiro Fukuhara and Takeshi Kurata.

ɹ The 17th Oriental International Committee for the Co-ordination and Standardization of Speech Databases and Assessment Techniques (O-COCOSDA2013), Nov. 2013.

10. Audio-Visual Voice Conversion using Noise-Robust Features.

ɹ Kohei Sawada, Masanori Takehara, Satoshi Tamura and Satoru Hayamizu.

ɹ 2014 IEEE International Conference on Acoustics, Speech and Signal Pro-cessing (ICASSP2014), pp.7949-7953, May 2014.

11. Improvement of utterance clustering by using employees^ʟsound and area data.

ɹ Tetsuya Kawase, Masanori Takehara, Satoshi Tamura, Satoru Hayamizu, Ryuhei Tenmoku and Takeshi Kurata.

ɹ 2014 IEEE International Conference on Acoustics, Speech and Signal Pro-cessing (ICASSP2014), pp.3071-3075, May 2014.

12. Spoken document retrieval using word co-occurrence information.

ɹ Kensuke Hara, Hiroaki Taguchi, Koudai Nakajima, Masanori Takehara, Satoshi Tamura and Satoru Hayamizu.

ɹ The 11th NTCIR Conference, Dec 2014.

13. Analysis of Customer Communication by Employee in Restaurant and Lead Time Estimation.

ɹ Masanori Takehara, Hiroya Nojiri, Satoshi Tamura, Satoru Hayamizu, Takeshi Kurata.

ɹ Asia-Pacific Signal and Information Processing Association Annual Sum-mit and Conference 2014 (APSIPA2014), Dec. 2014.

ߨԋʢݚڀձͱେձʣ

1. ࿑ಇू໿ܕαʔϏεʵैۀһߦಈܭଌٕज़ʹجͮ͘෼ੳͱՄࢹԽɽ ɹ ఱ໨ོฏɼ஛ݪਖ਼ۣɼ଎ਫޛɼଂా෢ࢤɽ

ɹ HCG^γϯϙδ΢Ϝ2010^࿦จूɼpp.443-448^ɼDec. 2010^ɽ

2. ೔ຊ৯Ϩετϥϯ࢈ۀʹ͓͚ΔϚϧνηϯαͱPOSσʔλʹجͮ͘αʔϏεΦϖ Ϩʔγϣϯਪఆɽ

ɹ ఱ໨ོฏɼ্Ԭྰࢠɼ຀ా޹࢚ɼ৽ଜ໠ɼ஛ݪਖ਼ۣɼ଎ਫޛɼଂా෢ࢤɽ ɹ ୈ10ճ৘ใՊֶٕज़ϑΥʔϥϜʢFIT2011ʣɼpp.859-860ɼSep. 2011ɽ

ドキュメント内 音声情報と行動計測による作業内容推定の研究 (ページ 40-68)