Enhanced Conditional Generative Adversarial Network For Handling Subject Variability In Human Activity Recognition
While splitting datasets, researchers assume that training set is exchangeable with test set and expect good classification performance. This assumption is invalid due to subject variability due to age differences. Classification models trained on activity data from one particular age group such as...
محفوظ في:
| المؤلف الرئيسي: | |
|---|---|
| التنسيق: | أطروحة |
| اللغة: | English |
| منشور في: |
2023
|
| الموضوعات: | |
| الوصول للمادة أونلاين: | http://eprints.usm.my/60094/1/ALI%20OLOW%20JIMALE%20-%20TESIS%20cut.pdf |
| الوسوم: |
إضافة وسم
لا توجد وسوم, كن أول من يضع وسما على هذه التسجيلة!
|
| الملخص: | While splitting datasets, researchers assume that training set is exchangeable with test set and expect good classification performance. This assumption is invalid due to subject variability due to age differences. Classification models trained on activity data from one particular age group such as adults cannot generalize to activity data collected from a different age group such as elderly. Subject variability in the context of age is a valid problem that degrades the performance of activity recognition, but remains unexplored. Existing studies that investigated subject variability in activity recognition overlooked this problem and only focused on contextual subject variability, and intra-subject variability. This study investigates the effects of subject variability on the performance of sensor-based activity recognition. Elderly and adult datasets were used to evaluate the assessment techniques. Experiments on adult dataset only, experiments on elderly dataset only, and experiments on adult (as training) and elderly (as test) datasets were conducted using machine learning and deep learning. The results show a significant performance drop in activity recognition on different subjects with different age groups. On average, the drop in recognition accuracy is 9.75% and 12% for machine learning and deep learning models respectively. Conditional Generative Adversarial Network (CGAN) is an ideal solution to address subject variability. |
|---|