Large-scale regression in the complex domain: Performance analysis of CRLS algorithms censoring noninformative data in an online manner

Mengüç E. C.

Niğde Ömer Halisdemir Üniversitesi Mühendislik Bilimleri Dergisi, vol.12, no.2, pp.349-359, 2023 (Peer-Reviewed Journal)


Extracting and learning meaningful information from big data streams paves the way for improving the quality of life of societies and the development of new technologies in the field of science and engineering. On the other hand, recent advances in sensor technology, increased availability of computing power and computer memory reveal that data is not just real-valued, but large-scale complex-valued datasets must also be dealt with. For this purpose, for the first time in this study, the performances of the recently proposed online censoring (OC) based complex-valued recursive least squares (OC-CRLS) and OC-based augmented CRLS (OC-ACRLS) algorithms are tested on large-scale regression problems and compared with those of their classical versions in the literature in detail. Simulation studies show that the OC-CRLS and OC-ACRLS algorithms significantly shorten the training time in large-scale regression problems defined in the complex domain without affecting testing performance in a negative way, due to the advantages of their OC mechanism. This proves that OC-CRLS and OC-ACRLS algorithms are effective and powerful algorithms in big data streaming applications that can be defined in the complex domain.