Infrared (IR) spectroscopy is a cornerstone analytical technique in gas logging, enabling the identification and quantification of hydrocarbons and non-hydrocarbons in geological formations during oil and gas exploration. The integration of machine learning (ML) techniques has revolutionized IR spectral analysis, offering enhanced accuracy, robustness, and automation over traditional methods. This review article provides a comprehensive examination of ML approaches, including regression models, neural networks, ensemble methods, and unsupervised learning, applied to quantitative analysis of IR spectra in gas logging. It evaluates their performance, adaptability to complex datasets, and ability to address challenges such as spectral overlap, noise, nonlinear relationships, and environmental variability. Current limitations, including data scarcity, model interpretability, and computational constraints, are discussed, alongside future research directions to optimize ML-driven IR spectral analysis for real-time field applications. This article aims to guide researchers and industry professionals in advancing ML-based gas logging solutions for the oil and gas industry.