walking-detection

• This is a project to demonstrate how to create "walking detection" models with various approaches.

  • By Kevin Chao (kevinchao@gmail.com)
  • https://www.linkedin.com/in/kevin-chao-com/
  • Latest updated on 2024-02-14

• Purpose:

  • The purpose of this project is to demonstrate how to create Machine Learning Models from raw accelerometer dataset. I use Walking Detection as an example.
  • The whole process aims to be simple and easy to understand the end-to-end pipeline of Data Science project.

• Data:

  • The raw accelerometer data was download from:
    • https://physionet.org/content/accelerometry-walk-climb-drive/1.0.0/
  • The sampling rate of the Acc data is 100 Hz
  • Data including labeled activities of walking, stair climbing, and driving.
  • Types of activities:
    • 1=walking (Walking)
    • 2=descending stairs (DescStairs)
    • 3=ascending stairs (AscendStairs)
    • 4=driving (Driving)
    • 77=clapping (Clapping)
    • 99=non-study activity (NonStudyAct)
  • Device positions:
    • lw: left wrist (WristL)
    • rw: right wrist (WristR)
    • lh: left hip (HipL)
    • rg: right hip (HipR)
    • la: left ankle (AnkleL)
    • ra: right ankle (AnkleR)
  • The unit of accelerometer is in g (9.8 m/s^2)

• Source:

  • 01_data_understanding.ipynb
    • Jupyter notebook to show how to understand the raw accelerometer data, conduct simple quality control, and show data visualization.
  • 02_data_preparation_time_series_analysis.ipynb
    • Show basic signal process steps to analize high-friquency accelerometer data.
    • Show algorithm to compute Step Count in both time and frequency domains
  • 03_data_preparation_features_creation.ipynb
    • Show how to create a dataframe with various features
  • 03b_data_preparation_features_creation_all_subjects.ipynb
    • Run all subjects' feature files
  • 04_modeling_evaluation.ipynb
    • Run Scikit-learn models and evaluate the model.
  • 04b_modeling_with_Spark.ipynb
  • 04c_modeling_with_PyTorch.ipynb
  • 04d_modeling_with_TensorFlow.ipynb

• Outputs:

  • Format of acc file:
    • Columns: 'subject_id', 'device_loc', 'act_id', 'act_name', 'event_num', 'walk_or_not', 'unique_id', 'time', 'acc_x', 'acc_y', 'acc_z'
    • Row: idf1ce9a0f, AnkleL, 1, Walking, 1, 1, idf1ce9a0f_AnkleL_1_Walking_1_1, 354.05, -0.070, -0.973, 0.078
    • .....
    • Row: idf1ce9a0f, AnkleL, 99, NonStudyAct, 547, 0, idf1ce9a0f_AnkleL_99_NonStudyAct_547_0, 3282.01, -0.059, 0.043, 0.992
  • ML modeling results will be placed under:
    • ~/outputs/xxML_outputs/

• Additions: