A new framework by using deep learning techniques for data processing

Deep auto-encoder neural networks have been widely used in several image classification and recognition problems, including handwriting recognition, medical imaging, face recognition, etc. The overall performance of deep auto-encoder neural networks mainly depends on the number of parameters used, structure of neural networks and the compatibility of the transfer functions. However, an inappropriate structure design can cause a reduction in the performance of deep auto-encoder neural networks. Four frameworks are proposed to evaluate the performance of the auto-encoder which is one of the common used deep learning techniques. In the first framework, the parameters of each auto-encoders were optimized by using Taguchi method. The proposed framework was validated by using four datasets; DDOS detection, IDS recognition, Epileptic seizure recognition and Digit classification datasets. In the second framework, pre-processing technique of energy spectral density was used to extract important features from input data. The proposed framework was tested by using three medical datasets. In the third framework, deep auto-encoder was combined with Discrete Wavelet Transform (DWT) to enhance its performance. Then, framework produced satisfactory results when compared to well-known studies in this field. Finally, a linear model was proposed as a post-processing technique to enhance the output of deep auto-encoder which its parameters were estimated by using Particle Swarm Optimization Algorithm (PSO). The experimental results show that the proposed method presented high acccuracy when compared with previous studies.