HUMAN ACTIVITY RECOGNITION FROM SMARTPHONES USING DATA CLASSIFICATION

K. Ammulu

Abstract


Activity recognition is one amongst  the foremost vital technology behind several applications like medical analysis, human survey system and  it is an active research  topic in health care and smart homes. Smart phones are equipped with numerous built-in sensing platforms like accelerometer , gyroscope, GPS, compass sensor and GPS sensor.  We can design a system to capture the state of the user. Activity recognition (AR) system takes the raw sensor reading from mobile sensors as inputs and estimates a human motion activity using data mining and machine learning techniques. In this paper, we analyze the performance of two classification algorithms i.e Random Forest (RF) and modified RF in an online activity recognition system working on Android platforms and this technique can supports on-line training and classification using the accelerometer data only. Usually first we use the RF classification algorithm program associated next we tend to utilize an improvement of modified random forest i.e MRF. For the purpose of activity recognition, modified RF will eliminates the computational complexities of RF by creating decision trees (creating smaller training sets for each actions and classification will be performed based on these reduced training sets). We will predict the performance of these classifiers from a series of observations on human activities like walking, running, lying down, sitting and standing in an online activity recognition system. In this paper, we are intended to analyze the performance of classifiers with limited training   data and limited  accessible memory on the phones compared  to off-line.

Keywords:  Activity Recognition, Classification, Random Forest, modified RF;


Full Text:

PDF

Refbacks

  • There are currently no refbacks.


© 2017 International Journal of Global Research in Computer Science (JGRCS)
Copyright Agreement & Authorship Responsibility