A Novel 1-D Convolution Neural Network With SVM Architecture for Real-Time Detection Applications

To enhance the performance and sensitivity of continuous monitoring systems for detection of chronic diseases, selection of optimal machine learning algorithms is pivotal. Presently, the commonly used algorithms face constraints, such as high computational cost and lack of optimal feature selection on application to real time signals thereby reducing the efficiency of such analysis. Deep learning approaches, such as the convolution neural network, overcome these drawbacks by calculating automated features from raw signal and classifying the derived features. This architecture shows good merits. However, the use of fully connected multi-layer perceptron algorithms have shown low classification performance. This paper proposes to develop a modified deep learning convolution neural network algorithm integrated with support vector machines to address the drawbacks present in multi-layer perceptron and thereby improving the overall performance of real-time detection applications. The system is validated on real-time breath signals for non-invasive detection of diabetes. The performance of this proposed algorithm is evaluated and compared with the existing technique.