AI-Driven Big Data Classification: A Taxonomy of Learning Paradigms, Scalability Approaches, and Emerging Research Directions
TL;DR
A taxonomy of existing methods into supervised and instance-based learning, deep and hybrid architectures, and imbalance-aware learning based on resampling is proposed, and a research-gap table identifies open issues in interpretability, cross-domain transfer, and energy-aware computation.
Abstract
The explosion of data from connected devices, digital platforms and sensor networks has highlighted the limitations of traditional classification algorithms and has driven the need for artificial intelligence (AI) methods that can scale to handle the volume, velocity and variety of data. This survey presents a comprehensive review of the current state of AI-enabled big data classification along three complementary axes: learning paradigms, scalability strategies and emerging research directions. We propose a taxonomy of existing methods into supervised and instance-based learning, deep and hybrid architectures, and imbalance-aware learning based on resampling.The infrastructural dimension of the problem is then discussed in relation to distributed computing frameworks such as MapReduce and Apache Spark MLlib, dimensionality-reduction pipelines and grid-based indexing schemes that keep classification tractable at scale. The surveyed application areas are healthcare and Internet of Things (IoT) systems, network intrusion detection, genomics, environmental monitoring, supply chains and multimodal emotion recognition. A comparison table compares representative techniques on method, domain and trade-offs, and a research-gap table identifies open issues in interpretability, cross-domain transfer, and energy-aware computation. The paper ends with a discussion of promising directions, including explainable hybrid pipelines, federated and streaming classification, and standardized benchmarking for large-scale artificial intelligence systems.