Abstract:
Human Activity Recognition (HAR) will become an essential part in activity
recognition for general medical and health care services. In this study, we have
designed and implemented an android application for the ease of caretakers which
aims to send an alert message to the attendee’s contact number about the Alzheimer’s
patient activity stats performed in a certain time interval. Using inbuilt sensors of
smart phones we are collecting patient’s data in our application and classifying it
using a supervised learning algorithm KNN. There are two actions that are selected
for recognition i.e. standing and walking. We have tried our best to optimize our
training model by collecting raw data in different scenarios as input and estimating a
patient’s activity using data mining concepts and machine learning techniques and
checking the accuracy by taking real time values to generate the best results for
prediction of an identified activity. During the research we analysed the performance
of four classification algorithms i.e. Naive Bayes (NB) , K Nearest Neighbour
(KNN), J48 and Random forests. As prediction results for KNN was achieved best
that’s why we have selected KNN as our classifier. Also we have observed the
change in the accuracy rates of prediction model by changing the value for K in
KNN. In this project we have intended to analyse the performance of classifier with
limited training data sets and limited set of activities also keeping in mind the limited
memory of phone storage. However, the accuracy of the system highly depend upon
the selected features and the quality oftraining model.
