Análisis predictivo y optimización estratégica: un enfoque empresarial para la toma de decisiones financieras

This systematization was developed in collaboration with RECREAMARKETING S.A., a company specialized in process automation, with the objective of implementing a predictive analytical model based on machine learning techniques to optimize commercial and financial decision-making using historical sales data. The CRISP-DM methodology was adopted, structuring the process into phases of business understanding, data preparation and transformation, modeling, training, validation, and deployment. The selected model was a multivariable LSTM neural network, configured to simultaneously predict five key variables (volume, net price, cost, tax, and price per liter). Data were segmented by SKU and distribution channel, and normalized to ensure quality and analytical compatibility. The model was trained in Python and further validated on the enterprise platform Watsonx.ai, showing a high degree of accuracy (R² > 0.97). Projections from April to December 2025 were consistent with historical trends and allowed the identification of strategic patterns, such as high-margin products and the most profitable channels. Results were presented through an interactive Power BI dashboard, consolidating key metrics, detailed analyses, and commercial optimization recommendations. The experience demonstrated the technical and strategic feasibility of using predictive models in data-driven business management, as well as the need to institutionalize data quality processes and evidence-based decision-making.

• Hovedforfatter: Mendoza Álava, Josselyn Nicolle (author)
• Format: bachelorThesis
• Sprog: spa
• Udgivet: 2025
• Fag: Análisis predictivo; Aprendizaje automático; Inteligencia artificial; LSTM
• Online adgang: http://repositorio.espam.edu.ec/handle/42000/2872