write a dissertation proposal for human activity Recognition using real world datasets from wearable sensor
the proposal outlines
Sensor based Human Activity Recognition
What is human activity recognition?
What Sensor based Human Activity Recognition
Research Context and Motivation
Lifestyles different these days
The need to recognize activity for more healthy life by knowing the METS “metabolic equivalent of task” and use t to calculate calories burns
Elderly people or patients who wants to life independent
Ethical people want to know about Mets activity and improve their health.
Research Objectives and Contributions
Need it to be more realistic and can be applying in real world scenarios
Most used datasets are controlled datasets which is not reflect the real-world situations
Most testing is done using intra-classification, we used inter-classification.
“Intra-subject accuracy, the datasets are randomly divided into training and testing sets to eliminate participant variation. This method is less practical and requires new participants to train models individually.
Inter-subject accuracy, the data of a single participant is excluded from the training dataset and included in the testing dataset; the rest participants’ data is used to train the classifier”
Employing the hybrid testing data “inserting a small representative quantity of training data provided by the test subject during the training phase and excluding it from the testing phase.”
Studding the effects and importuning of the features “demographic characteristics” Gender, age, weight, and so on.
The proposed research will be conducted in two phases
Build a model without using demographic characteristics and build a model with demographic characteristics and see the different in accuracy
For Build a model with demographic characteristics we will build: Heretical model multi-level classifier first recognize gender, age, then activity, and flat model “classic classifiers ex. Random Forest”. Include all features together
The performance of the proposed system will be evaluated using accuracy and F1-score metrics.
Showing the effects of including gender, weights in improving the recognition
please do not copy the same sentences in the outlines it is to use them to guide you.