Multidimensional Data Classification using Graph Signal Processing

Conventional classification methods encounter difficulties in handling non-linear relationships and high-dimensional feature spaces, highlighting the need for alternative approaches.Graph Signal Processing (GSP) has attracted many researchers the recent times owing to its ability to capture complexities associated with capturing intricate relationships within multidimensional data. In this paper, GSP is employed for the classification of digits. Graph signal is constructed from MNIST handwritten digit data. Fourier transformation is used to represent the signal variation on the graph obtained dataset.Leveraging the optimized graph signal regularization, the model captures intricate patterns, demonstrating superior adaptability for handwritten digit classification.A comparative analysis is performed on contemporary models such as Convolutional Neural Networks (CNN), Generative Adversarial Networks (GAN), and U-Net. The experimental results reveal that the proposed model based on GSP for classification is more efficient than the contemporary models for character recognition in terms of precision, recall and F-Score.