Insperied by the LSTMs for Human Activity Recognition and Deep Learning (and Machine Learning) for Human Activity Recognition repository.
Human Activity Recognition (HAR) using smartphones dataset and an LSTM. Classifying the type of movement amongst five categories:
WALKING
WALKING_UPSTAIRS
WALKING_DOWNSTAIRS
SITTING
STANDING
The sensor signals (accelerometer and gyroscope) were pre-processed by applying noise filters and then sampled in fixed-width sliding windows of 2.56 sec and 50% overlap (128 readings/window). The sensor acceleration signal, which has gravitational and body motion components, was separated using a Butterworth low-pass filter into body acceleration and gravity. The gravitational force is assumed to have only low frequency components, therefore a filter with 0.3 Hz cutoff frequency was used.
Skip this step if you want to use the existing data!
The file handel_data.ipynb is used to transform the HyperIMU CSV data to raw data with labels (first two boxes). Under comes the creation of the train and testing data.
Next open the file LSTM.ipynb. Run every box. Last one is starting to run the LSTM network. Output of actions will be written under the box and in the folder logs will .h5 files of the data, loss and accuracy plots, and comfusion matrix.
