Nowadays, companies and organizations have access to various data collection tools that enable them to amass vast amounts of data, which can be stored in databases. This data can be leveraged by machine learning algorithms to extract valuable information for decision-makers. However, this raw data is often of poor quality, containing errors such as missing data and outliers, requiring the intervention of technicians and domain specialists to prepare the data to ensure the \(F1\_Score \) of the analysis. This article proposes a framework for preparing high-quality data for machine learning algorithms, as manually identifying reliable data from a large pool can be challenging and time-consuming. Our approach is an architectural method that combines data preparation techniques to generate dataset quality.