Total: 1
This research project explores the optimization of the family selection process for participation in Uruguay's Crece Contigo Family Support Program (PAF) through machine learning. An anonymized database of 15,436 previous referral cases was analyzed, focusing on pregnant women and children under four years of age. The main objective was to develop a predictive algorithm capable of determining whether a family meets the conditions for acceptance into the program. The implementation of this model seeks to streamline the evaluation process and allow for more efficient resource allocation, allocating more team time to direct support. The study included an exhaustive data analysis and the implementation of various machine learning models, including Neural Networks (NN), XGBoost (XGB), LSTM, and ensemble models. Techniques to address class imbalance, such as SMOTE and RUS, were applied, as well as decision threshold optimization to improve prediction accuracy and balance. The results demonstrate the potential of these techniques for efficient classification of families requiring assistance.