Towards the Analysis of Regularized Denoising Autoencoder for Biosignal Processing: Lasso Versus Ridge Norms

The use of Internet of Things (IoT) that integrate smart bio sensor devices to the internet and shows individual health in real time. Healthcare organizations gain measurable insights into their most demanding problems like chronic diseases which demands long term monitoring. Various types of nano-sensors like Ingestible embedded in pills, Blood sampling sensor and tissue sensors are used in Healthcare IoT. Such implantable device collects, process and sends the vital signs called biosignals from particular organ of the human body where it has been implanted or fixed throughout the day to remote clinician. Such prolonged monitoring may weaken the battery power of nano-sensors. Since nano-sensors are miniaturized in nature and completely relies on its battery, energy awareness is incorporated in this paradigm that can help to avoid unnecessary energy consumption. This is achieved by data compression scheme. As the nano-sensors are light weight devices the designed algorithm should be low complex as well as efficient. As well, the signal acquired through this wireless sensor device are prone to be contaminated with noises because of the wearer’s movements. In this study, regularized denoising autoencoder (DAE) has been employed to compress and recover the signal from its noisy version. Lasso (Least Absolute Shrinkage and Selection Operator) and Ridge regularization concepts are used and contrasted in this article. The cost function now includes these penalty clauses to address the overfitting problem. The experimental findings demonstrate that LASSO Norm has outperformed over RIDGE in 18% for ECG, 57% for EMG & 31% for EEG signal with respect to Quality Score. The datasets used in this investigation were taken from a database that was open to the public for testing.