This project gives an explanation on how to build a Human activity Recognition (HAR) Model using Deep Learning.
Human Activity Recognition is a topic on which wide research is going on. These activities are recognized using the sensor readings. The main goal of this project is to train a Deep Learning model using these readings for this task.
The dataset used here is UCI HAR Dataset. This dataset can be accessed Here.
The dataset contains signals from the Gyroscope and Accelerometer of a smartphone while doing six activities : WALKING, WALKING_UPSTAIRS, WALKING_DOWNSTAIRS, SITTING, STANDING, LAYING.
For creating this dataset, experiments have been carried out with a group of 30 volunteers within an age bracket of 19-48 years. Each person performed six activities (WALKING, WALKING_UPSTAIRS, WALKING_DOWNSTAIRS, SITTING, STANDING, LAYING) wearing a smartphone (Samsung Galaxy S II) on the waist. Using its embedded accelerometer and gyroscope, we captured 3-axial linear acceleration and 3-axial angular velocity at a constant rate of 50Hz. The experiments have been video-recorded to label the data manually. The obtained dataset has been randomly partitioned into two sets, where 70% of the volunteers was selected for generating the training data and 30% the test data.
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. From each window, a vector of features was obtained by calculating variables from the time and frequency domain.
The dataset contains Time-Series data, therefore LSTM model was the best choice for this task.
The dataset contains ‘.txt’ files which contain signal data from the sensors. There are a total nine types of signals in the dataset : "body_acc_x","body_acc_y_","body_acc_z_","body_gyro_x_","body_gyro_y_","body_gyro_z_","total_acc_x_","total_acc_y_","total_acc_z_".-
These files were loaded in numpy arrays X_train, X_test, y_train and y_test. For this task, the functions ‘_get_x-’ and ‘get_y’were created.
The LSTM model is commonly utilized for Human Activity Recognition (HAR) due to its efficacy in handling time-series data. The model consists of four layers: LSTM Layer-1, Dropout Layer-1, LSTM Layer-2, Dropout Layer-2.
The Adam optimizer, known for its efficiency in optimization, was employed.
For multi-categorical classification, Softmax activation was utilized along with the "categorical_crossentropy" loss function, suitable for cases where the output variables (y_train, y_test) are one-hot encoded.
Dropout layers are strategically integrated to mitigate variance and prevent overfitting. After the model was compiled, various combinations of 'epochs' and 'batch_size' were experimented with to achieve the desired accuracy.
This file contains the code for creating the model. The path of the dataset needs to be changed according to you.
The model reached an impressive final accuracy of approximately 91%. This was achieved with a batch size of 32 and 20 epochs.
To tackle overfitting, a dropout probability of 0.7 was maintained in both dropout layers, effectively reducing variance.