The rapid proliferation of high-dimensional datasets across domains such as genomics, computer vision, cybersecurity, and Internet of Things (IoT) systems has intensified the challenges associated with the curse of dimensionality, including data sparsity, increased computational overhead, and degraded predictive generalization. As modern machine learning models increasingly rely on large-scale feature spaces derived from heterogeneous sensors, text corpora, and high-resolution imagery, the need for robust dimensionality mitigation strategies has become central to ensuring model reliability and scalability. This review systematically examines the methodological landscape of techniques designed to address these challenges, with particular emphasis on feature selection, dimensionality reduction, and representation learning frameworks. A structured review protocol inspired by the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) guidelines was adopted to identify and analyze peer-reviewed studies published between 2015 and 2025 from major scientific repositories. The selected literature encompasses experimental evaluations conducted on widely recognized benchmark datasets, including high-dimensional biomedical gene expression datasets, image recognition benchmarks such as CIFAR and MNIST, and large-scale text classification corpora. The review synthesizes algorithmic developments spanning classical statistical filters, wrapper-based optimization strategies, embedded learning models such as LASSO and Random Forests, and nonlinear transformation techniques including Principal Component Analysis (PCA), t-distributed Stochastic Neighbor Embedding (t-SNE), and deep autoencoder architectures. Comparative analysis of these approaches reveals consistent performance trade-offs between dimensionality reduction efficiency, computational complexity, and interpretability, particularly in scenarios involving noisy or highly correlated features. The findings highlight persistent limitations in scalability, real-time adaptability, and explainability of existing solutions, especially when deployed in dynamic, high-velocity data environments. By consolidating empirical evidence across diverse application contexts and identifying unresolved methodological constraints, this study provides a coherent reference framework for researchers and practitioners seeking to design more efficient high-dimensional learning pipelines. The principal contribution of this work lies in offering an integrative synthesis of contemporary dimensionality mitigation strategies, coupled with the identification of critical research gaps that can guide the development of next-generation machine learning systems.

Curse of Dimensionality High-Dimensional Data, Feature Selection, Dimensionality Reduction, Representation Learning, Machine Learning, Deep Learning, Data Preprocessing