Recurrence Plots and Cross Recurrence Plots

Lecture Notes in Networks and Systems, 613, 495–509p. (2023) DOI:10.1007/978-981-19-9379-4_36

Cancer Classification from High-Dimensional Multi-omics Data Using Convolutional Neural Networks, Recurrence Plots, and Wavelet-Based Image Fusion

High-dimensional and multi-modal data pose an exceptional challenge in machine learning. With the number of features vastly exceeding the number of training instances, such datasets often bring established pattern recognition techniques to an awkward position: Traditional, shallow models crumble under the sheer complexity of the data, but deep neural networks will helplessly overfit. In this study, an innovative methodology takes up the task, in the case study of using multi-modal biological data for binary classification of cancer types. Our deep learning approach entails transforming the data into images, integrating different modalities via wavelet-based image fusion, then extracting features, and classifying the data with pretrained convolutional neural networks. The results reveal that this framework has the potential to tackle high-dimensional data efficiently and effectively, learning from a low volume of complex data without overfitting, suggesting this to be a promising direction for further research.