Abstract

Marginalization in Mexico is a multidimensional phenomenon that persists even in cities with seemingly favorable socioeconomic indicators. Morelia, exemplifies this paradox: it reports a very low municipal marginalization index, yet exhibits high levels of labor precariousness and inequality. Traditional classification methods, based on descriptive statistics, sometimes overlook critical qualitative factors that define the real experience of marginalization. This study proposes a hybrid model combining Mamdani fuzzy logic to capture and consensus expert perception, and a neural network to predictively classify the degree of marginalization. Two hundred and sixty-seven (267) neighborhoods were analyzed based on five key dimensions aligned with CONEVAL and CONAPO, operationalized through 46 indicators evaluated by eight experts. Results indicate the model achieves high classification accuracy (85.9% in training), identifying basic services (100% normalized importance) and health (71.3%) as the most determining predictors of marginalization, unlike official metrics. The research concludes that this methodology offers a robust tool for identifying "invisible marginalization" and optimizing the allocation of public resources, directly aligning with Sustainable Development Goals 1 (end of poverty) and 10 (reduction of inequalities.

You must be a member to download the full article.