Application of Deep Neural Networks In the Field of Information Security and Healthcare
Computing Sciences and Computer Engineering
This work focuses on three different classification problems. The first is Intrusion Detection System (IDS) in network, the second is heart disease prediction and the last one is skin lesion classification in images using deep learning techniques. For IDS we have trained the deep neural network (DNN) model on KDD CUP 1999 data set. Similarly, for heart disease prediction we used the UCI Heart disease data set. The DNN is trained for 100 epochs. Furthermore, for skin lesion detection, in first stage pre-processing techniques are applied to remove noise and to improve visibility of lesion area. Then Preprocessed melanoma images of benign, malignant lesions are fed to the Convolutional Neural Network (CNN) learning model for classification. Different evaluation measures are applied to analyze results. Different techniques are applied to improve results such as stochastic gradient decent, augmentation and learning rate adjustment. The model is tested on 500 skin lesion images out of which 70% of the images are used for training and remaining 30% are used for testing purpose. Apart from this, in this work the Deep Neural Networks used for network intrusion detection and heart disease prediction has outperformed the state of the art method. Similarly, for CNN model our simulations results shows that our DNN model has good results in comparison to state of the art algorithm used for skin cancerous lesion detection.
(2019). Application of Deep Neural Networks In the Field of Information Security and Healthcare. 2019 SoutheastCon.
Available at: https://aquila.usm.edu/fac_pubs/18043