Decoding EEG During Action Observation, Motor Imagery, & Motor Execution

Evan Woods 2024-01-05

Support Vector Classifier Results

Motor Imagery While Sitting: Detection of Resting vs Action Observation

Subject 1:
     pred
truth  0  1
    0 10  2
    1  0 10
Accuracy: 90.909%

Subject 2:
     pred
truth  0  1
    0  5  3
    1  3 11
Accuracy: 72.727%

Subject 3:
     pred
truth  0  1
    0 12  0
    1  2  8
Accuracy: 90.909%

Subject 4:
     pred
truth  0  1
    0 11  0
    1  1 10
Accuracy: 95.455%

Subject 5:
     pred
truth  0  1
    0 13  1
    1  0  8
Accuracy: 95.455%

Subject 6:
     pred
truth  0  1
    0 10  2
    1  2  8
Accuracy: 81.818%

Subject 7:
     pred
truth  0  1
    0 10  0
    1  2 10
Accuracy: 90.909%

Subject 8:
     pred
truth  0  1
    0 10  3
    1  1  8
Accuracy: 81.818%

Mean Accuracy: 87.500%.
Standard Error: ±7.966%.

Motor Imagery While Sitting: Detection of Action Observation vs Motor Imagery

Subject 1:
     pred
truth  0  1
    0 13  0
    1  0  9
Accuracy: 100.000%

Subject 2:
     pred
truth  0  1
    0 11  0
    1  2  9
Accuracy: 90.909%

Subject 3:
     pred
truth  0  1
    0 11  1
    1  1  9
Accuracy: 90.909%

Subject 4:
     pred
truth  0  1
    0 13  0
    1  0  9
Accuracy: 100.000%

Subject 5:
     pred
truth  0  1
    0 14  0
    1  1  7
Accuracy: 95.455%

Subject 6:
     pred
truth  0  1
    0 12  2
    1  0  8
Accuracy: 90.909%

Subject 7:
     pred
truth  0  1
    0 10  0
    1  0 12
Accuracy: 100.000%

Subject 8:
     pred
truth  0  1
    0 13  2
    1  0  7
Accuracy: 90.909%

Mean Accuracy: 94.886%.
Standard Error: ±4.505%.

Motor Imagery While Standing: Detection of Resting vs Action Observation

Subject 1:
     pred
truth  0  1
    0 13  0
    1  0  9
Accuracy: 100.000%

Subject 2:
     pred
truth  0  1
    0  9  0
    1  1 12
Accuracy: 95.455%

Subject 3:
     pred
truth  0  1
    0 13  1
    1  0  8
Accuracy: 95.455%

Subject 4:
     pred
truth  0  1
    0 12  0
    1  0 10
Accuracy: 100.000%

Subject 5:
     pred
truth  0  1
    0 13  1
    1  1  7
Accuracy: 90.909%

Subject 6:
     pred
truth  0  1
    0 10  1
    1  0 11
Accuracy: 95.455%

Subject 7:
     pred
truth  0  1
    0 10  0
    1  0 12
Accuracy: 100.000%

Subject 8:
     pred
truth  0  1
    0 11  0
    1  0 11
Accuracy: 100.000%

Mean Accuracy: 97.159%.
Standard Error: ±3.382%.

Motor Imagery While Standing: Detection of Action Observation vs Motor Imagery

Subject 1:
     pred
truth  0  1
    0 13  0
    1  0  9
Accuracy: 100.000%

Subject 2:
     pred
truth  0  1
    0  8  1
    1  1 12
Accuracy: 90.909%

Subject 3:
     pred
truth  0  1
    0 13  1
    1  0  8
Accuracy: 95.455%

Subject 4:
     pred
truth  0  1
    0 11  1
    1  0 10
Accuracy: 95.455%

Subject 5:
     pred
truth  0  1
    0 13  1
    1  1  7
Accuracy: 90.909%

Subject 6:
     pred
truth  0  1
    0 10  1
    1  0 11
Accuracy: 95.455%

Subject 7:
     pred
truth  0  1
    0 10  0
    1  0 12
Accuracy: 100.000%

Subject 8:
     pred
truth  0  1
    0 11  0
    1  0 11
Accuracy: 100.000%

Mean Accuracy: 96.023%.
Standard Error: ±3.793%.

Comparision Against Results of Prior Research

The highest mean accuracy of the classifiers in the prior research is: 82.73%
with a standard error of ±2.54.

The lowest mean accuracy of the classifiers is: 87.500% with a standard error of
7.966%.

The highest mean accuracy of the classifiers is: 97.159% with a standard error
of 3.382%.

Increasing a Subject’s Model Accuracy

The lowest performing model of the Resting vs. Action Observation
classifications is the model for subject #2. The accuracy of subject #2's model
is: 72.727.

Explore the subject with the lowest performing model’s data and find outliers, high-leverage, or non-linearities.

Logistic Regression: Training & Validation

          truth
prediction  0  1
         0 16  4
         1  6 26

Validation Accuracy of Logistic Regression: 80.769%.

Detecting Outliers

There are no detected outliers in the logistic regression fit on the subject
with the lowest performing model's data.

Detecting and Removing High-Leverage Values

There are 16 high-leverage values:

      122        40        33       100        42        54        60       166 
0.1525843 0.1392962 0.1230248 0.2702669 0.1232194 0.1294163 0.1583125 0.1634064 
      291       190       298       273       209        75        47       105 
0.1968935 0.2429515 0.1299119 0.1517439 0.1382543 0.1825471 0.1501227 0.1632604 

Refitting a Logistic Regression Model

          truth
prediction  0  1
         0 17  2
         1  5 28

Validation Accuracy of Logistic Regression with no high leverage: 86.538%.

          truth
prediction  0  1
         0  5  3
         1  3 11

Accuracy of Logistic Regression on the subject with the lowest performing
model's Test Data after removing high-leverage: 72.727%. The previous accuracy
on test data with an SVM was: 72.727%.

Refitting the Lowest Performing Support Vector Classifier Model

          truth
prediction  0  1
         0  7  3
         1  1 11

[1] 81.81818

Results

The validation accuracy of the logistic regression model on the subject with
the lowest performing model's data increased model performance from 80.769% to
86.538% after removing high-leverage values detected in the subject's training
data.