Scalable Machine Learning Algorithms for Big Data–Driven Signal Analytics in Multi-Sensor Systems

Abstract

The extensive nature of heterogeneous multi- sensors systems has led to large amounts
of high dimensional signal data, which has become extremely demanding in issues of
scalability, fusion capability, and signal recovery performance. In a bid to solve these issues,
this paper puts forward a scaleable machine learning based framework of big data based
signal analytics in multi sensor environments. The suggested methodology is a convergence
of effective capacity of signal representation and multi-sensor fusion approach that is scaled
and a distributed learning architecture to permit effective signal augmentation, de-noising
and analytics in large-scale data circumstances. The common framework developed is a
system model to represent sensor heterogeneity, noise factors, scalability limitations, and
this is then followed by the creation of algorithmic system that will support distributed signal
processing yet will maintain signal fidelity. The validity of the given method is judged by
exhaustive experiments on multi- sensor data, and signal reconstruction measures of SNR,
MSE, PSNR and spectral distortion, as well as scalability measures of runtime, throughput and
computational efficiency. The experimental findings illustrate that the proposed framework
can always achieve better results with respect to reconstruction accuracy and scalability in
comparison with their traditional centralised and non-scalable baselines, especially when
the level of noise increases, as well as, when data size expands. These results support the
appropriateness of the suggested scalable learning structure in terms of high-performance
signal analytics with large-scale multi-sensors.