/detection_heavy__drinking_episodes

Primary LanguageJupyter Notebook

Detection of Heavy Drinking Episodes Using Smartphone Accelerometer Data

Details

  1. clean_tac: folder containing all the TAC readings for the 13 candidates
  2. all_accelerometer_data_pids_13: contains the tri-axial accelerometer data for 13 PIDs
  3. final_data.csv:
  • output of gen_features; contains all the features we extracted using the 2 window approach and summarizing statistics
  • 30874 datapoints x 180 features
  • this data is used by model_train to train the random forest classifier to get an accuracy of 82.27 %
  1. gen_features.py: python file to generate features from the raw data
  2. model_train: python file to use random forest classifier using different values of max depth ranging from 1 to 20 We have seen that for max_depth=20 the accuracy is the best so we ran it with a random seed fro 20 times and taken the average accuracy. It also shows a plot of how accuracy increases with max_depth.

To run any of the scripts - python3 <script.py>

References

Learning to Detect Heavy Drinking Episodes Using Smartphone Accelerometer Data

dataset