ICFES Prediction Project

In this work, a data science project focused on predictive evaluation of student performance in the Colombian State Education Examination (ICFES) is presented, employing supervised classification methodologies. The study encompassed an extensive dataset, including demographic information and socioeconomic characteristics of students, to predict academic success. The research was structured in stages of feature selection, modeling, model evaluation, analysis of the best model, and development of an API. Descriptions of previous processes such as univariate and multivariate analysis, imputation of missing data, and outlier detection were omitted, focusing directly on the mentioned phases. The use of performance metrics such as precision, recall, F1-score, and area under the ROC curve was highlighted to select the most appropriate model. The best model underwent a detailed analysis to understand its functioning and the implications of its predictions, which is essential for decision-making and the confidence of the end-users. Interpretation techniques were employed to clarify the contribution of each feature in the selected model. Finally, an API was developed using Flask, enabling real-time predictions. This project integrates the creation of applicable tools with academic analysis, promoting interaction and participation in the analysis of educational data, and facilitating informed decision-making in the educational sector.