Recently, machine learning (ML) has become a crucial tool for enhancing process quality control in manufacturing plants. However, real-time assessments are often challenging. Soft sensors, which can predict process quality indicators using ML, have gained significant attention since 2000 because of their advantages, such as process stability, reduced product rejection, and improved energy and fuel efficiency. Oil distillation, polymers, cement, and steel were the primary industries that developed soft sensors for quality indicators. Over time, more industries have adopted these models owing to the advantages previously mentioned. ML algorithms for processing soft sensors have evolved from simple linear algorithms to complex deep learning (DL) models with neural networks, support vector machines (SVMs), and tree-based models also being widely used. This article summarizes the methodologies implemented in soft-sensor technology during this century. To this end, a comprehensive selection of articles from different processes using ML algorithms was analyzed and discussed. As data availability and computing power increase, DL algorithms will become the primary focus of soft-sensor research, which will help lower energy consumption, enhance production rates, and reduce CO2 footprints.