Comparison of neural network back propagation algorithms for early detection of sleep disorders

Sleep is not merely a BREAK from our regular work. It is must to be physically and mentally refreshed every day. Having a sound nights sleep, one can perform best in whatever job in hand. But some time, sleep gets disturbed along with some awkward behaviors known as sleep disorders. The various techniques and practices are followed by numerous researchers for the diagnosis of the unusual behaviors which increase the disturbances in sleep and also encourage other sleep disorders. In this paper, a step has been taken towards the early detection of a few sleep disorders like Sleep Apnea, Insomnia, Parasomnia and Snoring using artificial neural network algorithms. The prior detection of these disorders can reduce the further effects on human body. This paper presents the comparison of four training algorithms gradient descent, quasi newton, conjugate gradient and Bayesian regularization by using different training functions such as trainrp, trainlm, trainscg and trainbr respectively. All these algorithms are trained by the data set acquired from various physicians. From the results, it is found that Bayesian regularization algorithm which is trained by using trainbr training function provides the best result for early detection of sleep disorders as per chosen sample size of 95 patient records.