23 August 2015

Purpose

Integrating and consolidating data from multiple sources.

Context

In this project, data for Human Activity Recognition1 (HAR) project from UCI Machine Learning Repository is downloaded, cleaned, and aggregated as a tidy data set. I have done this project as the course project for “Getting and Cleaning Data” in Data Science Specialization by John Hopkins University offered at Coursera.

Challenge

The data for this project is dispersed across multiple files:

  • Training Data
    • X_train.txt: Predictor feature values (normalized -1 to 1)
    • y_train.txt: Class labels (1 to 6)
    • subject_train.text: Subject identifier (1 to 30)
  • Test Data
    • X_test.txt: Predictor feature values (normalized -1 to 1)
    • y_test.txt: Class labels (1 to 6)
    • subject_test.text: Subject identifier (1 to 30)
  • Meta-data
    • features.txt: Feature names
    • activity_labels.txt: Activity labels

These files need to be merged and consolidated into a tidy dataframe before any meaningful analysis can be performed. Moreover, most data analysis, visualization, and machine learning libraries require data to be in a tidy dataframe.

dplyr library made it easy to select() the features needed, merge() (join) datasets on feature values, and calculating group summaries using group_by() and summarize_each().

The sensor data is finally averaged across subjects and activities and the data for mean and standard deviation for each activity are reported per subject.

Sensor Data

There are two types on sensors used in this experiement, accelerometer and gyroscope. Each produce 3-dimensional raw signals, tAcc-XYZ and tGyro-XYZ. Body and gravity acceleration signals are extracted from the raw signals. Then body linear acceleration and angular velocity are derived in time. Also the magnitude of these three-dimensional signals are calculated.

Finally a Fast Fourier Transform (FFT) was applied to some of these signals. The feature names and the transformations involved are detailed in the features_info file.

References

  1. Davide Anguita, Alessandro Ghio, Luca Oneto, Xavier Parra and Jorge L. Reyes-Ortiz. A Public Domain Dataset for Human Activity Recognition Using Smartphones. 21th European Symposium on Artificial Neural Networks, Computational Intelligence and Machine Learning, ESANN 2013. Bruges, Belgium 24-26 April 2013. 



blog comments powered by Disqus