Selecting the appropriate machine learning algorithm for various datasets in the social sciences can be a time-consuming process. In this work, we use supervised learning, specifically a Random Forest algorithm, to recommend the best algorithm for a given dataset based on dataset characteristics. To do this, we first describe each dataset using meta-features – more than one hundred meta-features are extracted from 50 datasets, from simple measures like the number of features, class distribution, and other statistics that capture the main characteristics of social science data from education and business domains. Using these meta-features on five machine learning algorithms, we train a Random Forest to predict which algorithm is likely to work best for new datasets. Our approach makes it easier for researchers to quickly select suitable models without manual trial and error. This approach can save time and help improve the quality of analysis in social science research.