Biometric_Activity_Classifier

Physical activity classification of biometric (IMU) data using machine learning
Inertial measurement unit (IMU) data consists of accelerometer and gyroscope on x, y, z axises

Neural Network

Confusion matrix: activity classifier from phone biometric data

Phone Neural Network Accuracy

Phone Neural Network Loss

SVM Results

Confusion matrix: activity classifier from phone biometric data

Confusion matrix: activity classifier from smartwatch biometric data

Analysis of phone biometric data

Accelerometer (m/s^2)

Scatter plot matrix

Line graph first minute

Gyroscope (radians/s)

Scatter plot matrix

Line graph first minute

Inertial measurement unit (IMU)

Scatter plot matrix

Line graph first minute

Notes:

Phone in pocket
Watch on wrist of dominant hand

20Hz sampling rate for phone and watch
~64800 samples -> ~54 minutes of data for each subject

Subject-id: unique to subject, Range: 1600-1650
ActivityLabel: unique activity, Range: A-S (no “N” value)
Timestamp: Integer, Linux time
x: x axis of sensor
y: y axis of sensor
z: z axis of sensor

Data Set Used

https://archive.ics.uci.edu/ml/datasets/WISDM+Smartphone+and+Smartwatch+Activity+and+Biometrics+Dataset+

References

Arduino Tiny ML tutorial for gesture recognition

https://colab.research.google.com/github/arduino/ArduinoTensorFlowLiteTutorials/blob/master/GestureToEmoji/arduino_tinyml_workshop.ipynb

https://blog.tensorflow.org/2019/11/how-to-get-started-with-machine.html

Helpful links

https://towardsdatascience.com/understanding-the-confusion-matrix-from-scikit-learn-c51d88929c79

https://en.wikipedia.org/wiki/Feature_scaling