mcboyd/userid-ml

User identification through activity classification from smartphone and smart-wearable sensor data using machine learning. Utilizing the "WISDM Smartphone and Smartwatch Activity and Biometrics Dataset".

UserID-ML

User identification through activity classification from smartphone and smart-wearable sensor data using machine learning. Utilizing the "WISDM Smartphone and Smartwatch Activity and Biometrics Dataset".

Data Extraction

Code, pseudocode, and concepts documented in data folder.

• Authorized Person: 90 seconds of data for each subject/activity is allocated to each of the testing and training data sets
• Imposter: 18 other subjects are randomly selected and 30 seconds of data for the same activity are randomly chosen for each subject
  • Data from 9 subjects is placed in the testing data set
  • While data from the other 9 is placed in the training data set
• Training class ratio is 90:270 (subject:imposter) by design

Random Forest Algorithm

Will use the TensorFlow class BoostedTreesClassifier as the model. Requirements:

• Separate features and labels with Dataframe.pop()
• Convert dataframe into Tensor with tf.data.Dataset.from_tensor_slices()
• Define input function to break data into batches
• Define evaluate function to run on test data
• Create model, train, and evaluate

Equal Error Rate (EER)

Pseudocode in equal_error_rate.py.

• Technical documentation and formulas: https://www.researchgate.net/post/Any_advice_on_computing_Equal_Error_Rate
• We can use ROC from the sklearn library
  • https://stackoverflow.com/questions/28339746/equal-error-rate-in-python
  • https://yangcha.github.io/EER-ROC/
• ROC (Receiver Operating Characteristic)
  • https://towardsdatascience.com/understanding-auc-roc-curve-68b2303cc9c5
  • https://www.statisticshowto.com/receiver-operating-characteristic-roc-curve/
• Plot the true positive rate against the false positive rate to get the ROC, then use the sklearn library to calculate the EER