1,a) 1,b) IoT SVM Random Forest GMM-HMM A Study on Data Analysis Aiming at Accuracy Improvement of In-Home Living Activity Akita Hiroya 1,a) Sato Kenya 1,b) 1. IoT(Internet of Things) [1][2] [3] IoT 1 Doshisha University Graduate School of Science and Engineering a) hiroya.akita@nislab.doshisha.ac.jp b) ksato@mail.doshisha.ac.jp 2. ECHONET Lite [4] Random Forest c 2017 Information Processing Society of Japan 1
6m 換気扇 机 パソコン 冷蔵庫 電子キッチンレンジ炊飯器 3m 1 ベッド 3 : 3. 2 3.1 1 2 3.2 1 1 iremocon ipod touch + テレビ机パソコンベッド エアコン 換気扇冷蔵庫 電子キッチンレンジ炊飯器 3m 4 : 2 iremocon[6] JSCA WG ECHONET-Lite ECHONET-Lite ECHONET-Lite ECHONET-Lite 2 1 3 2 4 BLE(Bluetooth Low Energy) [5] 4 beacon ipod touch ipod touch ipod touch 3.3 5 REST 2 3.4 c 2017 Information Processing Society of Japan 2
1 Rest サーバー mongodb -50 wifi の電波強度の遷移 温度湿度照度 位置情報加速度情報 -55-60 92 183 274 365 456 547 638 729 820 911 1002 1093 1184 1275 1366 1457 1548 1639 1730 1821 1912 2003 2094 2185 2276 2367 2458 2549 2640 2731 2822 2913 3004 3095 3186 3277 3368 3459 3550 3641 3732 3823 3914 4005 4096 4187 4278 4369 4460 4551-65 iremocon 温度, 湿度, 照度情報取得 ipod touch 位置情報加速度情報取得 電波 BLE 電波 電波強度 (dbm) -70-75 -80-85 5-90 2 OS ios9.3.5 ipod touch swift2.2 Xcode 7.3 2GB CPU mongodb 2.6.3 2 1 PC 4 4. 4.1 TV X,Y,Z BLE Wifi 2.4GHz Wifi 6 6 31 Wifi NaN 10-95 時間 (sec) 6 Wifi A F near middle far 1 4.2 PC 4.3 svm random forest python sklearn 4.3.1 SVM(Support Vector Machine) SVM 7 8 6 4 4 SVM 9 3 SVM F 8432 0.9716 0.9967 0.9840 4001 0.9680 0.9408 0.9542 3785 0.9986 1.0 0.9993 3856 0.9869 0.9980 0.9927 3074 0.9949 0.9714 0.9830 1523 0.9884 0.9868 0.9876 4002 0.9281 0.9100 0.9190 97.480% 4.3.2 Random Forest Random Forest c 2017 Information Processing Society of Japan 3
4 Random Forest F 8432 0.9844 0.9889 0.9867 4001 0.9624 0.9624 0.9624 3785 0.9912 1.0 0.9955 3856 0.9908 0.9914 0.9911 3074 0.9789 0.9797 0.9793 1523 0.9950 0.9934 0.9942 4002 0.9602 0.9421 0.9510 97.977% 7 SVM: 10 Rondom Forest: 8 SVM: SVM の主成分削減による正答率の遷移 100 80 正答率 (%) 60 40 20 0 25 16 15 14 13 12 11 10 9 8 7 6 5 4 3 2 1 主成分 ( 次元数 ) 11 Random Forest: 9 SVM: 10 11 6 4 4 Random Forest 12 4.3.3 7 c 2017 Information Processing Society of Japan 4
Random Forest の主成分削減による正答率の遷移 100 80 正答率 (%) 60 40 20 0 25 16 15 14 13 12 11 10 9 8 7 6 5 4 3 2 1 主成分 ( 次元数 ) 12 Random Forest: 14 Random Forest: 6 GMM-HMM 65.970% 10.182% 1.269% 95.222% 43.876% 97.689% 25.337% 13 SVM: PC SVM Random Forest 5 13 14 5 SVM 97.183% Random Forest 96.764% 4.3.4 GMM-HMM GMM-HMM GMM-HMM 3 6 5. 2 20 Wifi 10 Wifi 20 SVM 97% c 2017 Information Processing Society of Japan 5
Random Forest SVM SVM Random Forest Random Forest PC 3 SVM 84% Random Forest 93% SVM Random Forest SVM Random Forest IoT GMM-HMM GMM-HMM GMM-HMM JSPS 26540038 [1] S. Hongeng, R. Nevatia, and F. Bremond: Video-based event recognition: activity representation and probabilistic recognition methods, Computer Vision and Image Understanding, pp. 129162, (2004-11). [2] L. Ballan, M. Bertini, A. Del Bimbo, L. Seidenari, and G. Serra: Event detection and recognition for semantic annotation of video, Multimedia tools and applications, pp. 279302, (2011-1). [3] Ming-Je Tsai, Chao-Lin Wu, Sipun Kumar Pradhan, Yifei Xie, Ting-Ying Li, Li-Chen Fu, and Yi-Chong Zeng: Context-aware Activity Prediction using Human Behavior Pattern inreal Smart Home Environments, IEEE International Conference on Automation Science and Engineering (CASE), pp. 168-173, (2016-8). [4],,,,,,, : ECHONET Lite, (DICOMO2016), pp.1449-1457,(2016-7). [5] Aplix: MyBeacon, https://-store.stores.jp/,( 2017-7). [6] GLAMO INC: iremocon http://i-remocon.com/, ( 2017-7). 6. SVM Random Forest GMM-HMM c 2017 Information Processing Society of Japan 6