Semi-supervised learning exists in various real-world scenarios and can signifucantly impact scientific research in the field of biochemistry. There are also relevant specific applications in various fields, such as virus toxicity prediction, network security detection, application of soft sensors, etc. With the continuous breakthroughs in machine learning, there currently needs to be a complete review of research on semi-supervised learning methods and analyzes the challenges existing in the application process in this field; then, it sorts out and analyzes four methods of semi-supervised learning, including semi-supervised clustering, dimensionality reduction, regression, classification, and more advanced algorithms in these four different ways are written side by side. Then, the typical evaluation indicators of each algorithm (such as precision rate, recall rate, ROC curve, etc.) were introduced, and the effects of various semi-supervised learning algorithms were compared. The study found that semi-supervised learning methods are more accurate than fully supervised learning support vector machines rate, in which the SSC-EKE algorithm leads the traditional support vector machine classic supervised learning algorithm by absolute advantage. Finally, the practical application scenarios of semi-supervised learning are introduced, the future research directions of semi-supervised learning have prospected, and the full text is summarized.

Semi-supervised Learning, Semi-supervised Clustering, Semi-supervised Dimensionality Reduction, Semi-supervised Regression, Semi-supervised Classification, Evaluation Indicators