Sara Ashraf

Academic

Career

Verbatim text

AR Smarthomes 
 
Activity recognition is the prediction of an individual's daily life activities, usually indoors, based on 
ambient sensor data. Identifying a smart home occupant’s everyday activity, such as cooking a meal 
or watching TV, would enable elderly people to live independently in a safe and comfortable 
environment in their homes. A learning classifier's performance can be harmed by imbalanced 
activity instances of various classes within the dataset, as well as activities with fewer instances. We 
use deep learning techniques and generative adversarial networks in the proposed research to 
enhance the recognition efficiency of everyday activities in a smart home. The evaluation of the 
proposed approach on publicly available benchmark smart home datasets demonstrates its 
superior performance than existing techniques. 
Our goal is to apply existing Deep Learning techniques (DLT) to the problem of activity recognition 
in smart homes, with the main concern being an imbalanced dataset with some activities having 
more instances than others, as well as a small dataset with overlapping activities performed by 
multiple residents independently. For research purposes two deep learning models, CNN and LSTM 
were implemented on the smart home data, as well as Generative Adversarial Networks (GANs) 
technique to generate similar instances from a small dataset. 
 
 
 
 
 
 
 
 
 
 
 
 
Technology Used: 
Python, GitHub, Kaggle, Spyder  
Supervisor Name: 
Dr. Labiba Gillani Fahad 
Group Members:   
Sara Ashraf (i17 - 0285) 
Talha Nazir (i17 - 0324) 
Amna Zafar (i17 - 0325)