The tremendous amount of medical data leads to the early growth of applying machine learning models to clinical diagnostics. The problem then becomes how to more efficiently select relevant features from this immense and heterogeneous data. This chapter discusses the position of feature selection techniques used with medical data, more so filter, wrapper, and embedded methods, as they are an important step in increasing the precision, efficiency, and interpretability of diagnostic models. Feature selection focuses on methods that find the most informative attributes and remove the redundancy in order to alleviate the overfitting. The manuscript describes different feature selection methods along with their applications on clinical data, e.g., in electronic health records (EHRs) and genomics. In particular, the integration of these methods with AI and deep learning models is emphasized since feature selection has a significant impact on training and diagnostic outcomes of deep learning models. It also discusses challenges pertaining to dealing with high-dimensional data, missing values, and class imbalances, and provides practical solutions for the same. Additionally, the influence of robust feature selection in improving disease classification and prediction for oncology, cardiology, and neurology is illustrated through case studies. The chapter ends with expecting future developments of feature selection techniques along with AI technologies in order to provide even more precise, personalized medical care.