High-quality data is essential for the efficient functioning of medical AI models and significantly influences the accuracy and reliability of their predictions. Raw medical data often contains noise, inconsistencies, missing values, and biases that can dramatically impact model performance. This chapter discusses preprocessing methods for feature synthesis in healthcare, focusing on various data types, including tabulated medical data, temporal data, signal data, and imaging data systems. It includes basic preprocessing like data cleansing, data validation, data transformation, data normalization, and data augmentation techniques. It also outlines methods for imputing missing values, including statistical, machine learning, and deep learning approaches, as well as techniques for detecting outliers and removing noise. We cover transformation and normalization techniques, min–max scaling, z-score standardization, log transformations, box-cox transformations, and encoding methods. We further discussed data augmentation and feature generation techniques. The chapter also briefly covers the regulation and ethical considerations for medical data, as well as emerging trends and future directions. This chapter is a comprehensible summary of preprocessing methods that can be beneficial for feature synthesis improvement in AI-based medical applications.