Intelligent Multimetric Agglomerative Clustering for Robust Waveform Pattern Discovery in High Dimensional Signal Spaces

The waveform-based data is growing exponentially in the biomedical, communication, and industrial fields, which has generated an urgent need to develop unsupervised methods of learning the complex high-dimensional signal structures. Classical clustering techniques typically make use of one measure of similarity which does not reflect the inherent variability of waveforms patterns due to amplitude distortions, phase shifts and noise artifacts. In order to overcome such limitation, this paper will present an intelligent clustering framework called Intelligent Multimetric Agglomerative Clustering Network (IMAC-Net) which incorporates different distance metrics such as cosine, Euclidean, and cityblock into a single hierarchical structure. The proposed IMAC-Net is adaptive which benefits pattern discrimination in noisy signal conditions by weighting these metrics appropriately to balance between global and local similarity measures. Synthetic and real-world waveform dataset is used to verify the framework, and the performance is checked with the help of Silhouette Score, Davies-Bouldin Index, Calinski-Harabasz Score, and Cluster Purity. They are compared to the traditional methods, such as K-means, Hierarchical Clustering (single-metric), and Spectral Clustering. As experimental findings show, IMAC-Net exhibits better cluster cohesion, noise tolerance, and interpretation of waveform groups, which is much better as compared to the baseline models. The presented solution provides a generalizable and scalable solution to waveforms analytics, which opens the path to discovering patterns in high-dimensional, complex signal domains.