Human Activity Recognition Using Ambient Sensor Data

Variety and volume of data make human activity recognition especially interesting field for machine learning. It has thus seen incredible growth in past several years taking part in big data questions as well. In a broad sense question of HAR - human activity recognition is a very complex one, often times dealing with large amounts of data not belonging to a predefined class. However, this paper deals with supervised learning classifications task, focusing on several activity classes known as Activities of Daily Living - ADL. Generalized models for common activities and issues are looked into, and issues that appear due to the huge volume of data that is recognized as "other" when the models are applied to the real life data sets. Support vector machine method (SVM), naïve Bayes classifiers, KNN, Random Tree and Bagged Trees (Ensemble) algorithms are applied, and venturing into artificial neural network algorithms Multilayer Perceptron algorithm is applied as well.