Title: Supervised Deep Learning for Multi-Modal Big Data Fusion in IoT Systems: A Review
Authors: Virendra Tank
Volume: 10
Issue: 7
Pages: 165-172
Publication Date: 2026/07/28
Abstract:
The rapid proliferation of Internet of Things (IoT) devices has produced heterogeneous, high-volume, and high-velocity data streams spanning images, audio, inertial signals, environmental readings, and textual logs. Extracting reliable knowledge from such multi-modal big data requires fusion strategies that go beyond single-sensor analytics. Supervised deep learning has emerged as the dominant paradigm for this task because it can learn discriminative joint representations directly from labelled multi-modal streams without extensive hand-crafted feature engineering. This review synthesises the state of the art in supervised deep learning-based multi-modal data fusion for IoT systems. We organise the literature around a taxonomy of fusion levels (early, intermediate, and late fusion), summarise the deep architectures most frequently used for each level (convolutional neural networks, recurrent networks, autoencoders, and attention-based transformers), and examine representative application domains including human activity recognition, intrusion detection, industrial fault diagnosis, and health monitoring. We further present a comparative synthesis of reported performance trends across the reviewed studies, discuss recurring challenges such as modality heterogeneity, missing data, resource constraints, and interpretability, and outline promising directions including lightweight fusion for edge devices, self-supervised pretraining, and robustness to noisy or adversarial modalities. The review is based on 28 peer-reviewed and archival sources and is intended to orient researchers and practitioners entering this fast-growing field.