DS6999_HAR

Independent study on sampling rate and window size effect on human activity recognition model performance

Data files

Frequency Data: https://drive.google.com/open?id=1-vi2KLF_FhjBXSLqdTtb48EArqGHVJAN

• This is the raw data processed through get_data.py.
• It takes the raw data and applies all preliminary filters (median filter n=3, butterworth 0.3 Hz cutoff) and then downsamples it by taking every n-th reading.
• Downsample sets are 50 Hz (original), 25 Hz, 10 Hz, and 5 Hz.

Features Data: https://drive.google.com/open?id=1kki2UFE8WxfY-i_g2wXpwq60GZEPj8ue

• This is the Frequency data processed through feature_extract.py.
• It takes the frequency data and splits it into different observation windows (1s, 2s, 3s, 5s, and 10s).
• It calculates features from the time and frequency domain.

Training Data: https://drive.google.com/open?id=1_u81cMzNdJiTvHeI01k634HKwW8kvJ54

• This is the Features data processed through split_scale.py.
• This is the training data.
• It is preprocessed by removing all empty rows and empty/bad columns.
• The values are scaled between -1 and 1.

Test Data: https://drive.google.com/open?id=1Km_gLf2uo6jPu8VH_0isMLG31gxw-zS1

• This is the Features data processed through split_scale.py.
• This is the training data.
• It is preprocessed by removing all empty rows and empty/bad columns.
• The values are scaled between -1 and 1.

ML Data:

• This is the Test and Training data processed through different ML techniques.
• There are R and Python variants.

Writeup:

• This is the writeup for the findings of the data.

To-do list:

Apply ML techniques