際際滷

際際滷Share a Scribd company logo

Models Can Lie

Raju Rimal
Raju Rimal

An Illustration from Heart Rate Variability Data I have made an attempt to make calcification of different activities from Hear Rate Variability (HRV) data. I have also used model comparison here. Mostly it is graphics, but I will try to add some more text in future.

1 of 47
Models Can Lie An Illustration from Heart Rate Variablity Data Raju Rimal and Veronika Lindberg 2015/12/09 Norwegian University of Science and Technology Norwegian University of Life Sciences
Overview 1. Background 2. How data looks like 3. Classi鍖cation with series stack 4. Classi鍖cation with series averaged over Series repetition 5. Classi鍖cation with series averaged over Person-Event Combination 6. Some Comparison
Background
Some Background 1. PCR, PLS and Canonically Powered PLS (CPPLS) is used in the analysis
Some Background 1. PCR, PLS and Canonically Powered PLS (CPPLS) is used in the analysis 2. CPPLS integrate CCA with PLS to select relevant variables for response
Some Background 1. PCR, PLS and Canonically Powered PLS (CPPLS) is used in the analysis 2. CPPLS integrate CCA with PLS to select relevant variables for response 3. Cross-validation is performed over the observations on each a) frequency window b) series c) person-event combination
Some Background 1. PCR, PLS and Canonically Powered PLS (CPPLS) is used in the analysis 2. CPPLS integrate CCA with PLS to select relevant variables for response 3. Cross-validation is performed over the observations on each a) frequency window b) series c) person-event combination 4. Three variation of dataset is used
Some Background 1. PCR, PLS and Canonically Powered PLS (CPPLS) is used in the analysis 2. CPPLS integrate CCA with PLS to select relevant variables for response 3. Cross-validation is performed over the observations on each a) frequency window b) series c) person-event combination 4. Three variation of dataset is used Transpose of each frequency windows stacked together
Some Background 1. PCR, PLS and Canonically Powered PLS (CPPLS) is used in the analysis 2. CPPLS integrate CCA with PLS to select relevant variables for response 3. Cross-validation is performed over the observations on each a) frequency window b) series c) person-event combination 4. Three variation of dataset is used Transpose of each frequency windows stacked together The average frequencies over time for each Series
Some Background 1. PCR, PLS and Canonically Powered PLS (CPPLS) is used in the analysis 2. CPPLS integrate CCA with PLS to select relevant variables for response 3. Cross-validation is performed over the observations on each a) frequency window b) series c) person-event combination 4. Three variation of dataset is used Transpose of each frequency windows stacked together The average frequencies over time for each Series The average frequencies over time for each person-event combination
Some Background 1. PCR, PLS and Canonically Powered PLS (CPPLS) is used in the analysis 2. CPPLS integrate CCA with PLS to select relevant variables for response 3. Cross-validation is performed over the observations on each a) frequency window b) series c) person-event combination 4. Three variation of dataset is used Transpose of each frequency windows stacked together The average frequencies over time for each Series The average frequencies over time for each person-event combination 5. LDA model is used for discriminant analysis using the scores obtained form each of the latent variable model with cross-validation implemented
How data looks like
How data looks like 0 200 400 600 0 100 200 300 0 100 200 300 400 0 100 200 300 0 100 200 300 0 100 200 300 400 500 Person1Person1Person1Person1Person1Person3 P11P12P127P129P131P23 GymGymSaunaSaunaSaunaSauna 0.00 0.25 0.50 0.75 1.00 Frequency Window (dB) Time(s)
How data looks like 0 200 400 600 0 100 200 300 0 100 200 300 400 0 100 200 300 0 100 200 300 0 100 200 300 400 500 Person1Person1Person1Person1Person1Person3 P11P12P127P129P131P23 GymGymSaunaSaunaSaunaSauna 0.00 0.25 0.50 0.75 1.00 Frequency Window (dB) Time(s) Each block represent a series divided into several windows (rows), 128 columns each with 16 overlaps. The cell contains the frequency values obtained from fast fourier transform
How data looks like 0 200 400 600 0 100 200 300 0 100 200 300 400 0 100 200 300 0 100 200 300 0 100 200 300 400 500 Person1Person1Person1Person1Person1Person3 P11P12P127P129P131P23 GymGymSaunaSaunaSaunaSauna 0.00 0.25 0.50 0.75 1.00 Frequency Window (dB) Time(s) Each person may have involved into multiple activities which may have replications
How data looks like 0 200 400 600 0 100 200 300 0 100 200 300 400 0 100 200 300 0 100 200 300 0 100 200 300 400 500 Person1Person1Person1Person1Person1Person3 P11P12P127P129P131P23 GymGymSaunaSaunaSaunaSauna 0.00 0.25 0.50 0.75 1.00 Frequency Window (dB) Time(s) Set 1: Each windows are stacked in a row to form a big matrix (may su鍖er from repeated measurement). This contains di鍖erent parts of same series in various rows.
How data looks like 0 200 400 600 0 100 200 300 0 100 200 300 400 0 100 200 300 0 100 200 300 0 100 200 300 400 500 Person1Person1Person1Person1Person1Person3 P11P12P127P129P131P23 GymGymSaunaSaunaSaunaSauna 0.00 0.25 0.50 0.75 1.00 Frequency Window (dB) Time(s) Set 2: Each series are averaged over di鍖erent time points. Each row corresponds to one series.
How data looks like 0 200 400 600 0 100 200 300 0 100 200 300 400 0 100 200 300 0 100 200 300 0 100 200 300 400 500 Person1Person1Person1Person1Person1Person3 P11P12P127P129P131P23 GymGymSaunaSaunaSaunaSauna 0.00 0.25 0.50 0.75 1.00 Frequency Window (dB) Time(s) Set 3: A person can have multiple series of same activity (replication), the third set is averaged over each person-event conbination. In this case each row corresponds to some speci鍖c event for some speci鍖c person
Cross-Validation j cvTest 1, . . . j 1, j + 1, . . . , 10 cvTrain
Cross-Validation j cvTest 1, . . . j 1, j + 1, . . . , 10 cvTrain MVR models (PCR, PLS, CPPLS)
Cross-Validation j cvTest 1, . . . j 1, j + 1, . . . , 10 cvTrain MVR models (PCR, PLS, CPPLS) 1, 2, . . . , 128 Scores
Cross-Validation j cvTest 1, . . . j 1, j + 1, . . . , 10 cvTrain MVR models (PCR, PLS, CPPLS) 1, 2, . . . , 128 Scores y categories, X Scores from 1 to i LDA Model: y f (X) Loop over each additional scores
Cross-Validation j cvTest 1, . . . j 1, j + 1, . . . , 10 cvTrain MVR models (PCR, PLS, CPPLS) 1, 2, . . . , 128 Scores y categories, X Scores from 1 to i LDA Model: y f (X) Loop over each additional scores Error (cvTest and Train) For ith Scores and jth cv split Cross-validation Loop
Classification with series stack
Training and Cross-validation Errors PLS PCR CPLS Comp: 4 Error: 0.89 Comp: 79 Error: 0.35 Comp: 56 Error: 0.82 Comp: 128 Error: 0.3 Comp: 29 Error: 0.81 Comp: 70 Error: 0.35 0.84 0.88 0.92 0.3 0.4 0.5 0.6 0.7 0.8 0.9 cvtesttrain 0 50 100 0 50 100 0 50 100 Components MisclassificationError
Misclassi鍖cations Training Misclassi鍖cations PLS PCR CPLS 124 154 151 99 890 240 4 1 890 37 7 251 28 153 95 226 1948 420 60 50 291 1271 698 74 155 123 100 150 872 20 4 4 1 79 10 17 27 214 65 161 87 264 1910 37 5 4 2 3 2 203 1359 574 198 148 130 86 164 84 5 19 5 1 4 75 14 13 31 196 83 158 90 258 1916 34 8 1 5 1 4 194 1368 565 207 Light Jogging Walking Walking In Stairs Alcohol Cold Shower Cycling Gym Gym Restitution Morning Before Exercise Morning Test Sauna Sauna Ice Bath Shower Sleeping Unknown 0.00 0.25 0.50 0.75 1.00 0.00 0.25 0.50 0.75 1.00 0.00 0.25 0.50 0.75 1.00 Correct Predictions FALSE TRUE
Misclassi鍖cations Test Misclassi鍖cation PLS PCR CPLS 8 27 36 17 21 370 9 26 36 9 29 34 3 11 24 36 9 29 32 5 Light Jogging Resting Walking Walking In Stairs 0.00 0.25 0.50 0.75 1.00 0.00 0.25 0.50 0.75 1.00 0.00 0.25 0.50 0.75 1.00 Correct Predictions FALSE TRUE
Plotting Scores Scoreplot for PCR model event Alcohol Cold Shower Cycling Gym Gym Restitution Light Jogging Morning Before Exercise Morning Test Sauna Sauna Ice Bath Shower Sleeping Unknown Walking Walking In Stairs 4 2 0 2 30 20 10 0 10 Comp1 (59.39%) Comp2(7.38%) Model:PCR 4 2 0 2 4 6 4 2 0 2 Comp2 (7.38%) Comp3(4.31%) Model:PCR
Plotting Scores Scoreplot for PLS model event Alcohol Cold Shower Cycling Gym Gym Restitution Light Jogging Morning Before Exercise Morning Test Sauna Sauna Ice Bath Shower Sleeping Unknown Walking Walking In Stairs 4 2 0 2 20 10 0 10 Comp1 (59.27%) Comp2(7.34%) Model:PLS 4 2 0 2 4 2 0 2 Comp2 (7.34%) Comp3(2.31%) Model:PLS
Plotting Scores Scoreplot for CPPLS model event Alcohol Cold Shower Cycling Gym Gym Restitution Light Jogging Morning Before Exercise Morning Test Sauna Sauna Ice Bath Shower Sleeping Unknown Walking Walking In Stairs 3 2 1 0 1 2 4 2 0 2 Comp1 (8.13%) Comp2(20.65%) Model:CPPLS 2 0 2 4 6 3 2 1 0 1 2 Comp2 (20.65%) Comp3(3.89%) Model:CPPLS
Classification with series aver- aged over Series repetition
Training and Cross-validation Errors PLS PCR CPLS Comp: 15 Error: 0.98 Comp: 114 Error: 0.01 Comp: 128 Error: 0.99 Comp: 127 Error: 0.02 Comp: 68 Error: 0.99 Comp: 99 Error: 0.01 0.985 0.990 0.995 1.000 0.05 0.10 0.15 0.20 cvtesttrain 0 50 100 0 50 100 0 50 100 Components MisclassificationError
Misclassi鍖cations Training Misclassi鍖cations PLS PCR CPLS 1 3 1 3 30 0 3 0 1 0 1 0 3 0 1 0 1 3 174 2 4 0 1 0 2 0 1 8 12 1 3 1 3 0 3 0 3 0 1 0 1 0 3 0 1 0 1 0 177 0 6 0 1 0 2 0 1 2 18 1 3 1 3 0 3 0 3 0 1 0 1 0 3 0 1 0 1 0 177 0 6 0 1 0 2 0 1 3 17 Light Jogging Walking Walking In Stairs Alcohol Cold Shower Cycling Gym Gym Restitution Morning Before Exercise Morning Test Sauna Sauna Ice Bath Shower Sleeping Unknown 0.00 0.25 0.50 0.75 1.00 0.00 0.25 0.50 0.75 1.00 0.00 0.25 0.50 0.75 1.00 Correct Predictions FALSE TRUE
Misclassi鍖cations Test Misclassi鍖cation PLS PCR CPLS 0 2 2 0 2 20 0 2 2 0 2 1 1 0 2 2 0 2 1 1 Light Jogging Resting Walking Walking In Stairs 0.00 0.25 0.50 0.75 1.00 0.00 0.25 0.50 0.75 1.00 0.00 0.25 0.50 0.75 1.00 Correct Predictions FALSE TRUE
Plotting Scores Scoreplot for PCR model event Alcohol Cold Shower Cycling Gym Gym Restitution Light Jogging Morning Before Exercise Morning Test Sauna Sauna Ice Bath Shower Sleeping Unknown Walking Walking In Stairs 6 4 2 0 10 5 0 5 10 Comp1 (74.44%) Comp2(9.41%) Model:PCR 3 2 1 0 1 2 6 4 2 0 Comp2 (9.41%) Comp3(3.83%) Model:PCR
Plotting Scores Scoreplot for PLS model event Alcohol Cold Shower Cycling Gym Gym Restitution Light Jogging Morning Before Exercise Morning Test Sauna Sauna Ice Bath Shower Sleeping Unknown Walking Walking In Stairs 6 4 2 0 2 5 0 5 Comp1 (63.41%) Comp2(20.2%) Model:PLS 2 0 2 6 4 2 0 2 Comp2 (20.2%) Comp3(4.03%) Model:PLS
Plotting Scores Scoreplot for CPPLS model event Alcohol Cold Shower Cycling Gym Gym Restitution Light Jogging Morning Before Exercise Morning Test Sauna Sauna Ice Bath Shower Sleeping Unknown Walking Walking In Stairs 1 0 1 4 3 2 1 0 1 Comp1 (15.39%) Comp2(1.79%) Model:CPPLS 2 1 0 1 1 0 1 Comp2 (1.79%) Comp3(2.21%) Model:CPPLS
Classification with series aver- aged over Person-Event Combina- tion
Training and Cross-validation Errors PLS PCR CPLS Comp: 1 Error: 1 Comp: 5 Error: 0.26 Comp: 1 Error: 1 Comp: 5 Error: 0.38 Comp: 1 Error: 1 Comp: 1 Error: 0.87 0.50 0.75 1.00 1.25 1.50 0.4 0.6 0.8 cvtesttrain 1 2 3 4 5 1 2 3 4 5 1 2 3 4 5 Components MisclassificationError
Misclassi鍖cations Training Misclassi鍖cations PLS PCR CPLS 20 20 30 0 1 0 1 10 10 10 10 10 1 2 10 0 1 10 1 3 20 20 30 0 1 0 1 10 10 10 10 10 1 2 10 0 1 10 1 3 20 20 0 3 10 10 10 10 10 10 10 30 10 10 10 40 Light Jogging Walking Walking In Stairs Alcohol Cold Shower Cycling Gym Gym Restitution Morning Before Exercise Morning Test Sauna Sauna Ice Bath Shower Sleeping Unknown 0.00 0.25 0.50 0.75 1.00 0.00 0.25 0.50 0.75 1.00 0.00 0.25 0.50 0.75 1.00 Correct Predictions FALSE TRUE
Misclassi鍖cations Test Misclassi鍖cation PLS PCR CPLS 20 2 20 20 20 2 20 20 20 2 20 0 2 Light Jogging Resting Walking Walking In Stairs 0.00 0.25 0.50 0.75 1.00 0.00 0.25 0.50 0.75 1.00 0.00 0.25 0.50 0.75 1.00 Correct Predictions FALSE TRUE
Plotting Scores Scoreplot for PCR model event Alcohol Cold Shower Cycling Gym Gym Restitution Light Jogging Morning Before Exercise Morning Test Sauna Sauna Ice Bath Shower Sleeping Unknown Walking Walking In Stairs 4 2 0 2 10 5 0 5 Comp1 (72.28%) Comp2(13.79%) Model:PCR 2 0 2 4 2 0 2 Comp2 (13.79%) Comp3(7.15%) Model:PCR
Plotting Scores Scoreplot for PLS model event Alcohol Cold Shower Cycling Gym Gym Restitution Light Jogging Morning Before Exercise Morning Test Sauna Sauna Ice Bath Shower Sleeping Unknown Walking Walking In Stairs 4 2 0 2 5 0 5 10 Comp1 (72.27%) Comp2(13.72%) Model:PLS 2 0 2 4 2 0 2 Comp2 (13.72%) Comp3(7.12%) Model:PLS
Plotting Scores Scoreplot for CPPLS model event Alcohol Cold Shower Cycling Gym Gym Restitution Light Jogging Morning Before Exercise Morning Test Sauna Sauna Ice Bath Shower Sleeping Unknown Walking Walking In Stairs 0.4 0.2 0.0 0.2 0.4 0.4 0.2 0.0 0.2 Comp1 (2.32%) Comp2(2.57%) Model:CPPLS 0.25 0.00 0.25 0.50 0.4 0.2 0.0 0.2 0.4 Comp2 (2.57%) Comp3(0.26%) Model:CPPLS
Some Comparison
Misclassi鍖cation Errors Set 1 Set 2 Set 3 29 56 4 68 128 15 1 1 1 0.0175 0.0219 0.0789 0.3124 0.3276 0.3733 0.6667 0.875 0.9315 0.9521 0.9932 1 CPLS PCR PLS CPLS PCR PLS CPLS PCR PLS Model MisclassificationError Test Train Figure: Training and Test Misclassi鍖cation Error for all the three models. The LDA models were 鍖tted with the scores obtained from three models with components (number above each points) needed to get minimum cross-validation error.
References [1] M Dowle et al. data.table: Extension of Data.frame. R package version 1.9.6. 2015. url: http://CRAN.R-project.org/package=data.table. [2] Ulf G Indahl, Kristian Hovde Liland, and Tormod N脱s. Canonical partial least squaresa uni鍖ed PLS approach to classi鍖cation and regression problems. In: Journal of Chemometrics 23.9 (2009), pp. 495504. [3] Uwe Ligges, Tom Short, and Paul Kienzle. signal: Signal Processing. R package version 0.7-6. 2015. url: http://CRAN.R-project.org/package=signal. [4] Harald Martens and Magni Martens. Multivariate analysis of quality: an introduction. John Wiley & Sons, 2001. [5] Harald Martens and Tormod Naes. Multivariate calibration. John Wiley & Sons, 1992. [6] Hadley Wickham. ggplot: An Implementation of the Grammar of Graphics. In: R package version 0.4. 0 (2006). [7] Hadley Wickham. reshape2: Flexibly reshape data: a reboot of the reshape package. In: R package version 1.2 (2012). [8] Yihui Xie. knitr: A General-Purpose Package for Dynamic Report Generation in R. R package version 1.11. 2015. url: http://CRAN.R-project.org/package=knitr.

More Related Content

Models Can Lie

  • 1. Models Can Lie An Illustration from Heart Rate Variablity Data Raju Rimal and Veronika Lindberg 2015/12/09 Norwegian University of Science and Technology Norwegian University of Life Sciences
  • 2. Overview 1. Background 2. How data looks like 3. Classi鍖cation with series stack 4. Classi鍖cation with series averaged over Series repetition 5. Classi鍖cation with series averaged over Person-Event Combination 6. Some Comparison
  • 4. Some Background 1. PCR, PLS and Canonically Powered PLS (CPPLS) is used in the analysis
  • 5. Some Background 1. PCR, PLS and Canonically Powered PLS (CPPLS) is used in the analysis 2. CPPLS integrate CCA with PLS to select relevant variables for response
  • 6. Some Background 1. PCR, PLS and Canonically Powered PLS (CPPLS) is used in the analysis 2. CPPLS integrate CCA with PLS to select relevant variables for response 3. Cross-validation is performed over the observations on each a) frequency window b) series c) person-event combination
  • 7. Some Background 1. PCR, PLS and Canonically Powered PLS (CPPLS) is used in the analysis 2. CPPLS integrate CCA with PLS to select relevant variables for response 3. Cross-validation is performed over the observations on each a) frequency window b) series c) person-event combination 4. Three variation of dataset is used
  • 8. Some Background 1. PCR, PLS and Canonically Powered PLS (CPPLS) is used in the analysis 2. CPPLS integrate CCA with PLS to select relevant variables for response 3. Cross-validation is performed over the observations on each a) frequency window b) series c) person-event combination 4. Three variation of dataset is used Transpose of each frequency windows stacked together
  • 9. Some Background 1. PCR, PLS and Canonically Powered PLS (CPPLS) is used in the analysis 2. CPPLS integrate CCA with PLS to select relevant variables for response 3. Cross-validation is performed over the observations on each a) frequency window b) series c) person-event combination 4. Three variation of dataset is used Transpose of each frequency windows stacked together The average frequencies over time for each Series
  • 10. Some Background 1. PCR, PLS and Canonically Powered PLS (CPPLS) is used in the analysis 2. CPPLS integrate CCA with PLS to select relevant variables for response 3. Cross-validation is performed over the observations on each a) frequency window b) series c) person-event combination 4. Three variation of dataset is used Transpose of each frequency windows stacked together The average frequencies over time for each Series The average frequencies over time for each person-event combination
  • 11. Some Background 1. PCR, PLS and Canonically Powered PLS (CPPLS) is used in the analysis 2. CPPLS integrate CCA with PLS to select relevant variables for response 3. Cross-validation is performed over the observations on each a) frequency window b) series c) person-event combination 4. Three variation of dataset is used Transpose of each frequency windows stacked together The average frequencies over time for each Series The average frequencies over time for each person-event combination 5. LDA model is used for discriminant analysis using the scores obtained form each of the latent variable model with cross-validation implemented
  • 13. How data looks like 0 200 400 600 0 100 200 300 0 100 200 300 400 0 100 200 300 0 100 200 300 0 100 200 300 400 500 Person1Person1Person1Person1Person1Person3 P11P12P127P129P131P23 GymGymSaunaSaunaSaunaSauna 0.00 0.25 0.50 0.75 1.00 Frequency Window (dB) Time(s)
  • 14. How data looks like 0 200 400 600 0 100 200 300 0 100 200 300 400 0 100 200 300 0 100 200 300 0 100 200 300 400 500 Person1Person1Person1Person1Person1Person3 P11P12P127P129P131P23 GymGymSaunaSaunaSaunaSauna 0.00 0.25 0.50 0.75 1.00 Frequency Window (dB) Time(s) Each block represent a series divided into several windows (rows), 128 columns each with 16 overlaps. The cell contains the frequency values obtained from fast fourier transform
  • 15. How data looks like 0 200 400 600 0 100 200 300 0 100 200 300 400 0 100 200 300 0 100 200 300 0 100 200 300 400 500 Person1Person1Person1Person1Person1Person3 P11P12P127P129P131P23 GymGymSaunaSaunaSaunaSauna 0.00 0.25 0.50 0.75 1.00 Frequency Window (dB) Time(s) Each person may have involved into multiple activities which may have replications
  • 16. How data looks like 0 200 400 600 0 100 200 300 0 100 200 300 400 0 100 200 300 0 100 200 300 0 100 200 300 400 500 Person1Person1Person1Person1Person1Person3 P11P12P127P129P131P23 GymGymSaunaSaunaSaunaSauna 0.00 0.25 0.50 0.75 1.00 Frequency Window (dB) Time(s) Set 1: Each windows are stacked in a row to form a big matrix (may su鍖er from repeated measurement). This contains di鍖erent parts of same series in various rows.
  • 17. How data looks like 0 200 400 600 0 100 200 300 0 100 200 300 400 0 100 200 300 0 100 200 300 0 100 200 300 400 500 Person1Person1Person1Person1Person1Person3 P11P12P127P129P131P23 GymGymSaunaSaunaSaunaSauna 0.00 0.25 0.50 0.75 1.00 Frequency Window (dB) Time(s) Set 2: Each series are averaged over di鍖erent time points. Each row corresponds to one series.
  • 18. How data looks like 0 200 400 600 0 100 200 300 0 100 200 300 400 0 100 200 300 0 100 200 300 0 100 200 300 400 500 Person1Person1Person1Person1Person1Person3 P11P12P127P129P131P23 GymGymSaunaSaunaSaunaSauna 0.00 0.25 0.50 0.75 1.00 Frequency Window (dB) Time(s) Set 3: A person can have multiple series of same activity (replication), the third set is averaged over each person-event conbination. In this case each row corresponds to some speci鍖c event for some speci鍖c person
  • 19. Cross-Validation j cvTest 1, . . . j 1, j + 1, . . . , 10 cvTrain
  • 20. Cross-Validation j cvTest 1, . . . j 1, j + 1, . . . , 10 cvTrain MVR models (PCR, PLS, CPPLS)
  • 21. Cross-Validation j cvTest 1, . . . j 1, j + 1, . . . , 10 cvTrain MVR models (PCR, PLS, CPPLS) 1, 2, . . . , 128 Scores
  • 22. Cross-Validation j cvTest 1, . . . j 1, j + 1, . . . , 10 cvTrain MVR models (PCR, PLS, CPPLS) 1, 2, . . . , 128 Scores y categories, X Scores from 1 to i LDA Model: y f (X) Loop over each additional scores
  • 23. Cross-Validation j cvTest 1, . . . j 1, j + 1, . . . , 10 cvTrain MVR models (PCR, PLS, CPPLS) 1, 2, . . . , 128 Scores y categories, X Scores from 1 to i LDA Model: y f (X) Loop over each additional scores Error (cvTest and Train) For ith Scores and jth cv split Cross-validation Loop
  • 25. Training and Cross-validation Errors PLS PCR CPLS Comp: 4 Error: 0.89 Comp: 79 Error: 0.35 Comp: 56 Error: 0.82 Comp: 128 Error: 0.3 Comp: 29 Error: 0.81 Comp: 70 Error: 0.35 0.84 0.88 0.92 0.3 0.4 0.5 0.6 0.7 0.8 0.9 cvtesttrain 0 50 100 0 50 100 0 50 100 Components MisclassificationError
  • 26. Misclassi鍖cations Training Misclassi鍖cations PLS PCR CPLS 124 154 151 99 890 240 4 1 890 37 7 251 28 153 95 226 1948 420 60 50 291 1271 698 74 155 123 100 150 872 20 4 4 1 79 10 17 27 214 65 161 87 264 1910 37 5 4 2 3 2 203 1359 574 198 148 130 86 164 84 5 19 5 1 4 75 14 13 31 196 83 158 90 258 1916 34 8 1 5 1 4 194 1368 565 207 Light Jogging Walking Walking In Stairs Alcohol Cold Shower Cycling Gym Gym Restitution Morning Before Exercise Morning Test Sauna Sauna Ice Bath Shower Sleeping Unknown 0.00 0.25 0.50 0.75 1.00 0.00 0.25 0.50 0.75 1.00 0.00 0.25 0.50 0.75 1.00 Correct Predictions FALSE TRUE
  • 27. Misclassi鍖cations Test Misclassi鍖cation PLS PCR CPLS 8 27 36 17 21 370 9 26 36 9 29 34 3 11 24 36 9 29 32 5 Light Jogging Resting Walking Walking In Stairs 0.00 0.25 0.50 0.75 1.00 0.00 0.25 0.50 0.75 1.00 0.00 0.25 0.50 0.75 1.00 Correct Predictions FALSE TRUE
  • 28. Plotting Scores Scoreplot for PCR model event Alcohol Cold Shower Cycling Gym Gym Restitution Light Jogging Morning Before Exercise Morning Test Sauna Sauna Ice Bath Shower Sleeping Unknown Walking Walking In Stairs 4 2 0 2 30 20 10 0 10 Comp1 (59.39%) Comp2(7.38%) Model:PCR 4 2 0 2 4 6 4 2 0 2 Comp2 (7.38%) Comp3(4.31%) Model:PCR
  • 29. Plotting Scores Scoreplot for PLS model event Alcohol Cold Shower Cycling Gym Gym Restitution Light Jogging Morning Before Exercise Morning Test Sauna Sauna Ice Bath Shower Sleeping Unknown Walking Walking In Stairs 4 2 0 2 20 10 0 10 Comp1 (59.27%) Comp2(7.34%) Model:PLS 4 2 0 2 4 2 0 2 Comp2 (7.34%) Comp3(2.31%) Model:PLS
  • 30. Plotting Scores Scoreplot for CPPLS model event Alcohol Cold Shower Cycling Gym Gym Restitution Light Jogging Morning Before Exercise Morning Test Sauna Sauna Ice Bath Shower Sleeping Unknown Walking Walking In Stairs 3 2 1 0 1 2 4 2 0 2 Comp1 (8.13%) Comp2(20.65%) Model:CPPLS 2 0 2 4 6 3 2 1 0 1 2 Comp2 (20.65%) Comp3(3.89%) Model:CPPLS
  • 31. Classification with series aver- aged over Series repetition
  • 32. Training and Cross-validation Errors PLS PCR CPLS Comp: 15 Error: 0.98 Comp: 114 Error: 0.01 Comp: 128 Error: 0.99 Comp: 127 Error: 0.02 Comp: 68 Error: 0.99 Comp: 99 Error: 0.01 0.985 0.990 0.995 1.000 0.05 0.10 0.15 0.20 cvtesttrain 0 50 100 0 50 100 0 50 100 Components MisclassificationError
  • 33. Misclassi鍖cations Training Misclassi鍖cations PLS PCR CPLS 1 3 1 3 30 0 3 0 1 0 1 0 3 0 1 0 1 3 174 2 4 0 1 0 2 0 1 8 12 1 3 1 3 0 3 0 3 0 1 0 1 0 3 0 1 0 1 0 177 0 6 0 1 0 2 0 1 2 18 1 3 1 3 0 3 0 3 0 1 0 1 0 3 0 1 0 1 0 177 0 6 0 1 0 2 0 1 3 17 Light Jogging Walking Walking In Stairs Alcohol Cold Shower Cycling Gym Gym Restitution Morning Before Exercise Morning Test Sauna Sauna Ice Bath Shower Sleeping Unknown 0.00 0.25 0.50 0.75 1.00 0.00 0.25 0.50 0.75 1.00 0.00 0.25 0.50 0.75 1.00 Correct Predictions FALSE TRUE
  • 34. Misclassi鍖cations Test Misclassi鍖cation PLS PCR CPLS 0 2 2 0 2 20 0 2 2 0 2 1 1 0 2 2 0 2 1 1 Light Jogging Resting Walking Walking In Stairs 0.00 0.25 0.50 0.75 1.00 0.00 0.25 0.50 0.75 1.00 0.00 0.25 0.50 0.75 1.00 Correct Predictions FALSE TRUE
  • 35. Plotting Scores Scoreplot for PCR model event Alcohol Cold Shower Cycling Gym Gym Restitution Light Jogging Morning Before Exercise Morning Test Sauna Sauna Ice Bath Shower Sleeping Unknown Walking Walking In Stairs 6 4 2 0 10 5 0 5 10 Comp1 (74.44%) Comp2(9.41%) Model:PCR 3 2 1 0 1 2 6 4 2 0 Comp2 (9.41%) Comp3(3.83%) Model:PCR
  • 36. Plotting Scores Scoreplot for PLS model event Alcohol Cold Shower Cycling Gym Gym Restitution Light Jogging Morning Before Exercise Morning Test Sauna Sauna Ice Bath Shower Sleeping Unknown Walking Walking In Stairs 6 4 2 0 2 5 0 5 Comp1 (63.41%) Comp2(20.2%) Model:PLS 2 0 2 6 4 2 0 2 Comp2 (20.2%) Comp3(4.03%) Model:PLS
  • 37. Plotting Scores Scoreplot for CPPLS model event Alcohol Cold Shower Cycling Gym Gym Restitution Light Jogging Morning Before Exercise Morning Test Sauna Sauna Ice Bath Shower Sleeping Unknown Walking Walking In Stairs 1 0 1 4 3 2 1 0 1 Comp1 (15.39%) Comp2(1.79%) Model:CPPLS 2 1 0 1 1 0 1 Comp2 (1.79%) Comp3(2.21%) Model:CPPLS
  • 38. Classification with series aver- aged over Person-Event Combina- tion
  • 39. Training and Cross-validation Errors PLS PCR CPLS Comp: 1 Error: 1 Comp: 5 Error: 0.26 Comp: 1 Error: 1 Comp: 5 Error: 0.38 Comp: 1 Error: 1 Comp: 1 Error: 0.87 0.50 0.75 1.00 1.25 1.50 0.4 0.6 0.8 cvtesttrain 1 2 3 4 5 1 2 3 4 5 1 2 3 4 5 Components MisclassificationError
  • 40. Misclassi鍖cations Training Misclassi鍖cations PLS PCR CPLS 20 20 30 0 1 0 1 10 10 10 10 10 1 2 10 0 1 10 1 3 20 20 30 0 1 0 1 10 10 10 10 10 1 2 10 0 1 10 1 3 20 20 0 3 10 10 10 10 10 10 10 30 10 10 10 40 Light Jogging Walking Walking In Stairs Alcohol Cold Shower Cycling Gym Gym Restitution Morning Before Exercise Morning Test Sauna Sauna Ice Bath Shower Sleeping Unknown 0.00 0.25 0.50 0.75 1.00 0.00 0.25 0.50 0.75 1.00 0.00 0.25 0.50 0.75 1.00 Correct Predictions FALSE TRUE
  • 41. Misclassi鍖cations Test Misclassi鍖cation PLS PCR CPLS 20 2 20 20 20 2 20 20 20 2 20 0 2 Light Jogging Resting Walking Walking In Stairs 0.00 0.25 0.50 0.75 1.00 0.00 0.25 0.50 0.75 1.00 0.00 0.25 0.50 0.75 1.00 Correct Predictions FALSE TRUE
  • 42. Plotting Scores Scoreplot for PCR model event Alcohol Cold Shower Cycling Gym Gym Restitution Light Jogging Morning Before Exercise Morning Test Sauna Sauna Ice Bath Shower Sleeping Unknown Walking Walking In Stairs 4 2 0 2 10 5 0 5 Comp1 (72.28%) Comp2(13.79%) Model:PCR 2 0 2 4 2 0 2 Comp2 (13.79%) Comp3(7.15%) Model:PCR
  • 43. Plotting Scores Scoreplot for PLS model event Alcohol Cold Shower Cycling Gym Gym Restitution Light Jogging Morning Before Exercise Morning Test Sauna Sauna Ice Bath Shower Sleeping Unknown Walking Walking In Stairs 4 2 0 2 5 0 5 10 Comp1 (72.27%) Comp2(13.72%) Model:PLS 2 0 2 4 2 0 2 Comp2 (13.72%) Comp3(7.12%) Model:PLS
  • 44. Plotting Scores Scoreplot for CPPLS model event Alcohol Cold Shower Cycling Gym Gym Restitution Light Jogging Morning Before Exercise Morning Test Sauna Sauna Ice Bath Shower Sleeping Unknown Walking Walking In Stairs 0.4 0.2 0.0 0.2 0.4 0.4 0.2 0.0 0.2 Comp1 (2.32%) Comp2(2.57%) Model:CPPLS 0.25 0.00 0.25 0.50 0.4 0.2 0.0 0.2 0.4 Comp2 (2.57%) Comp3(0.26%) Model:CPPLS
  • 46. Misclassi鍖cation Errors Set 1 Set 2 Set 3 29 56 4 68 128 15 1 1 1 0.0175 0.0219 0.0789 0.3124 0.3276 0.3733 0.6667 0.875 0.9315 0.9521 0.9932 1 CPLS PCR PLS CPLS PCR PLS CPLS PCR PLS Model MisclassificationError Test Train Figure: Training and Test Misclassi鍖cation Error for all the three models. The LDA models were 鍖tted with the scores obtained from three models with components (number above each points) needed to get minimum cross-validation error.
  • 47. References [1] M Dowle et al. data.table: Extension of Data.frame. R package version 1.9.6. 2015. url: http://CRAN.R-project.org/package=data.table. [2] Ulf G Indahl, Kristian Hovde Liland, and Tormod N脱s. Canonical partial least squaresa uni鍖ed PLS approach to classi鍖cation and regression problems. In: Journal of Chemometrics 23.9 (2009), pp. 495504. [3] Uwe Ligges, Tom Short, and Paul Kienzle. signal: Signal Processing. R package version 0.7-6. 2015. url: http://CRAN.R-project.org/package=signal. [4] Harald Martens and Magni Martens. Multivariate analysis of quality: an introduction. John Wiley & Sons, 2001. [5] Harald Martens and Tormod Naes. Multivariate calibration. John Wiley & Sons, 1992. [6] Hadley Wickham. ggplot: An Implementation of the Grammar of Graphics. In: R package version 0.4. 0 (2006). [7] Hadley Wickham. reshape2: Flexibly reshape data: a reboot of the reshape package. In: R package version 1.2 (2012). [8] Yihui Xie. knitr: A General-Purpose Package for Dynamic Report Generation in R. R package version 1.11. 2015. url: http://CRAN.R-project.org/package=knitr.