This paper delivers a strategy to build a deep neural network, established by heaping layers of autoencoder, which in turn consists of both encoder and decoder layers, which are generally being locally trained to denoise the corrupted inputs and reconstruct an approximation to the original input. The outcome as an algorithm is a candid variation by stacking the ordinary autoencoder. It is basically a classification problem of machine learning yielding to obtain less classification error, and therefore spanning the performance gap with deep belief neural networks and in majority of the cases surpassing it. Results show that the reconstruction of the inputs depend upon the training parameters such as the upsurge of the epoch and batch size will increase the training period, thus increasing the accuracy in representing the denoised reconstruction.