Maestría Análisis de Datos y Sistemas Inteligentes

URI permanente para esta colecciónhttp://hdl.handle.net/11634/72664

Examinar

Mostrando 1 - 11 de 11

Construcción de un Cerebro Analítico en la Nube: Un Sistema de Inteligencia Artificial para la Predicción de Riesgos Hidrológicos en las Cuencas del AMB
(Universidad Santo Tomás, 2026-09-06) Mora Rivera, Edgar Mauricio; Calderón Castro, Wilson Fabián; Núñez Rodríguez, Rafael Augusto; Díaz Jaimes, Pedro Pablo; Isaza Lancheros, Maria Camila; Universidsad Santo Tomás
Hydrometeorological risk management in water utility infrastructure requires a transition from closed, local Supervisory Control and Data Acquisition (SCADA) monitoring schemes toward distributed architectures that integrate cloud computing and artificial intelligence. This master's thesis consolidates the development, implementation, and results of a hydrometeorological monitoring ecosystem for the Acueducto Metropolitano de Bucaramanga (amb), Colombia, comprising four components: (a) statistical characterization and early-warning threshold definition for the Arnania and Florida-Frío watersheds, using percentiles (P90, P95, P99) computed over 10-to-22-year historical series of daily precipitation; (b) optimization of a VBScript ingestion middleware running on the AVEVA Edge SCADA platform, which reduced worst-case blocking time from 180 to 24 seconds (an 86.7% improvement) by eliminating six simultaneous HTTP connections; (c) deployment of an interactive dashboard (Streamlit on Google Cloud Run, backed by BigQuery) for real-time visualization of six data sources; and (d) integration of a conversational artificial intelligence agent (Dialogflow CX) that answers natural-language queries about hydrometeorological variables with geographic context and operational recommendations. As an additional contribution, operational percentiles for four gauges (Martín Gil, La Mariana, Tona, and El Pajal) were independently recalculated from raw consolidated series, showing reasonable agreement with current thresholds in most cases and localized divergences —particularly at the 99th percentile— attributable to differences in the cut-off period, which points to the need for periodic recalibration. The results show that the convergence of statistical analytics, Edge-to-Cloud architecture, and conversational artificial intelligence is a viable, low-operating-cost mechanism to strengthen decision-making in water resource management.
Diagnóstico del registro administrativo censo arbóreo de Bucaramanga
(Universidad Santo Tomás, 2026-09-07) Benitez Goyeneche, Maria Alejandra; Martínez Reyes, Diana Carolina; Niño Sandoval, Kelly Johanna, Kelly; Universidad Santo Tomás
The Bucaramanga mayor's office has among its functions the generation of high-quality, consistent, and coherent data for the production of national statistics; in this case, a diagnostic assessment is carried out on the administrative record of Bucaramanga's tree census, which contains 71,586 records and 45 variables that must be evaluated using the methodology established by the National Department of Statistics (DANE). The following document describes the review of the database, the completion of the "ficha-revisión-base-de-datos" (database review form) through an automated process in Microsoft Excel, and the calculation of six quality indicators (variable type, length, allowed values, completeness, domain, and thematic validation rules). The completeness (99%) and variable type correspondence (93%) indicators received excellent and good ratings, respectively; field length reached 88% (good); while the allowed values (74%), domain (74%), and validation rules (77%) indicators showed poor and acceptable performance. The findings show that the registry is structurally suitable for statistical use, but it requires standardizing variables such as fecha (date), n_arbol (tree number), and comuna (district), and implementing data capture validations to strengthen data quality for the 2026 census update.
DataPulse Solutions: Creación de una Empresa de Analítica de Datos e Inteligencia de Negocios Orientada a Pequeñas y Medianas Empresas en Colombia
(Universidad Santo Tomás, 2026-09-04) Noriega Polania, Andres Eduardo; Sánchez Gonzalez, Pedro Luis; Mesa Martínez, Ana Carolina; Martínez Reyes, Diana Carolina; Universidad Santo Tomás
This work presents the business creation proposal for DataPulse Solutions, a firm devoted to democratizing access to data analytics and business intelligence for Colombian small and medium-sized enterprises (SMEs). The objective is to comprehensively design the business model —value proposition, market study, organizational structure, and technical and financial feasibility— and to demonstrate its viability through a functional prototype. The document is organized into five chapters: industry contextualization, problem statement and value proposition, market study and business model, organizational structure and feasibility, and conclusions with growth projections. The methodology follows a mixed approach grounded in state-of-the-art review, technology surveillance, gap analysis, and business model design through the Business Model Canvas, supported by official secondary sources (MinTIC, DANE, Confecámaras, Gartner) and complemented by the development and validation of a working prototype. Findings reveal a structural gap: Colombia has 850,000 formally registered SMEs, yet only 18% employ analytical tools for decision-making. The serviceable available market amounts to COP 518.5 billion. An analytical diagnostic agent was technically validated, developed with Streamlit, pandas, Plotly, Prophet, and GPT-OSS 120B, and executed on 10,194 real retail transactions from the Superstore dataset, achieving 86.3% sales forecasting accuracy at zero operating cost. The discussion examines the root causes of the gap cost barriers, talent shortage, and limited managerial awareness and contrasts the proposal against existing offerings designed for corporate budgets. The study concludes that the model is viable and sustainable without external capital: it requires an initial investment of COP 6,793,208, reaches break-even between months 2 and 9, and projects revenues of COP 595,000,000 by the third year, representing 0.11% of the available market.
Prototipo de asistente operativo inteligente (DABU) integrado con herramientas SCADA (Rockwell) para apoyo en la operación de estaciones de producción de crudo
(Universidad Santo Tomás, 2026-09-04) Buitrago Murillo, Daniel Alcides; Dallos Ramírez, Andrés Felipe; Núñez Rodríguez, Rafael Augusto; Peña Galvis, Omar Leonardo; Universidad Santo Tomás
The fragmentation of operational data in oil production facilities represents a critical bottleneck for timely decision-making within Centralized Control Rooms (CCR). This research presents the design, development, and implementation of DABU (Intelligent Operational Assistant), a hybrid neuro-symbolic platform integrated with industrial SCADA systems under Rockwell Automation's PlantPAx distributed architecture. The system is deployed across a six-layer virtualized environment (ISA-95) using independent virtual machines, enforcing a zero-trust network topology. DABU's cognitive engine combines the precision of a deterministic Python execution unit, which evaluates telemetry thresholds and alarms in real time, with a Retrieval-Augmented Generation (RAG) pipeline for unstructured technical manuals (PDFs) and historical logs (Excel sheets) indexed in ChromaDB. Conversation orchestration is governed by state graphs in LangGraph, routing language synthesis to a quantized local Large Language Model (phi3:mini) operating on-premise without reliance on external cloud APIs. Experimental validation in an emulated industrial environment confirmed DABU's ability to streamline operational triage and deliver context-grounded technical diagnostics. No hallucinations were observed in deterministic calculations during the evaluated trials, supporting the robustness of the proposed hybrid architecture and demonstrating its suitability as a resilient and model-agnostic solution for critical OT infrastructure.
Aplicación de modelos de machine learning para predicciones de exportaciones para el mercado floricultor colombiano
(Universidad Santo Tomás, 2026-09-01) Puentes Abril, Felipe; Becerra Angarita, Oscar Fernando; Valencía Niño, Cesar Hernando; Universidad Santo Tomás
This applied research project implemented, evaluated, and compared two machine learning models –Random Forest and XGBoost– to predict the monthly volume of Colombian flower exports to its three main destination markets: the United States, the United Kingdom, and Canada. The study integrated macroeconomic variables from Colombia and each importing country (gross domestic product, inflation, and exchange rate), together with seasonality and time-lag variables, over a destination-country/month panel of 396 observations covering the 2015–2025 period. The development was carried out in Python on Google Colab, following a methodology that included cross-validation for hyperparameter optimization, chronological data partitioning, and formal statistical validation through the Wilcoxon signed-rank test. Results show that XGBoost achieved a slightly better point performance than Random Forest on three of the four evaluated metrics (MAE, RMSE, and R²), although the Wilcoxon signed-rank test, applied to a 72 observation test set, confirmed that this difference is not statistically significant. Variableimportance analysis, complemented with SHAP values, showed a statistically significant agreement (ρ = 0.663; p = 0.0135) that both models rank the importing market’s economic size and demand seasonality as the most explanatory factors, while Colombia’s own macroeconomic conditions carry comparatively less weight. The project demonstrates the technical feasibility of applying ensemble learning algorithms to an agricultural foreigntrade problem, and its main contribution is a complete, reproducible methodology that can be reused by PROCOLOMBIA, Colombia’s Ministry of Commerce, Industry and Tourism, or by future research within the Colombian flower-growing sector.
Predicción temprana de brotes de dengue en Bucaramanga mediante un modelo de Machine Learning con integración de variables climáticas, geográficas y sociodemográficas
(Universidad Santo Tomás, 2026-08-28) Araque Villamizar, Reinaldo; Paipa Bautista, Diego Alberto; Álvarez Pizarro, Yuli Andrea; Universidad Santo Tomás
Problem. Dengue is an endemic arboviral disease in Bucaramanga that accumulated 10,259 cases reported to the National Public Health Surveillance System (SIVIGILA) between 2015 and 2022, with a hospitalization rate of 26.6% and recurrent epidemic peaks. However, the municipality lacks tools capable of anticipating outbreaks. Objective. To develop a machine learning model for the early prediction of dengue outbreaks in Bucaramanga by integrating epidemiological, climatic, geographic, and sociodemographic variables. Methods. A quantitative, applied, and longitudinal study was conducted following the CRISP-DM methodology. Data from SIVIGILA, climate records from IDEAM, Sentinel-2 vegetation indices (NDVI), and sociodemographic indicators from DANE were integrated and aggregated by epidemiological week and administrative district. Five approaches were compared: a seasonal naïve model and ARIMA as baseline models, versus Random Forest, XGBoost, and LSTM. Models were evaluated using a strict temporal split and prediction horizons ranging from one to four weeks. Performance was assessed using MAE, RMSE, MAPE, MASE, F1-score, and AUC metrics, complemented by the Diebold–Mariano test and SHAP-based interpretability analysis. Results. Machine learning models reduced prediction error to less than two-thirds of the naïve forecast and significantly discriminated outbreak weeks (p < 0.001). Temperature, with a seven-week lag, was identified as the most relevant environmental predictor. Outbreak risk was concentrated in the San Francisco, Centro, and Oriental districts, which accounted for 32.4% of reported cases. Conclusions. Dengue outbreaks can be anticipated using publicly available local data. The DengueAlert-BGA prototype represents an interpretable and replicable tool to strengthen epidemiological surveillance and public health response in the municipality.
Desarrollo de un ecosistema integrado de gestión y visualización de datos que centraliza y consolida los indicadores de resultados del Plan de Desarrollo Municipal, el Índice de Capacidad Estadística Territorial y la Rendición de cuentas de Niños, Niñas y Adolescentes, en la Alcaldía de Bucaramanga
(Universidad Santo Tomás, 2026-08-28) Soto Garnica, Yordy Andres; Pérez Acero, Cristian Leonardo; Moreno Granados, Brandon Esneyder; Pimentel Díaz, Carlos Daniel; Medina Monterrosa, Prudencia; Universidad Santo Tomás
This report documents the professional internship carried out at the Municipal Government of Bucaramanga. The institution previously lacked an interoperable architecture, relying on records scattered across departments and manual processes for handling its indicators. This disconnection delayed the issuance of reports and hindered decision-making. The main objective was to develop an integrated data management and visualization ecosystem to centralize the indicators of the Municipal Development Plan (PDM), the Territorial Statistical Capacity Index (ICET) and the Accountability Report on Children and Adolescents (RCNNA).Methodologically, a three-tier architecture was implemented. The data tier consolidated the information into a master file; the logic tier automated cleaning and transformation (ETL) through Power Query; and the presentation tier was deployed in Power BI, structuring the evaluation logic with DAX measures and the visual dashboard.As project results, 215 prioritized indicators were consolidated and 145 of them were technically characterized through metadata sheets. The Power BI dashboard was structured with four navigation levels: cover page, prioritized indicator batteries, individual baseline indicator detail, and historical time-series analysis. In addition, a Control and Efficiency Panel was developed to audit integrated data, result variation, and server update and synchronization times. It is concluded that the ecosystem successfully integrated the information. As a future scaling plan, migration to a relational engine (Azure SQL) is recommended, connecting data directly from each department's source systems and eliminating manual intervention.
Validación de Soft Sensing con Redes Neuronales e Híbridas y Prescripción en Sistemas de Decisión para la Gestión del Riesgo Nitrogenado en Acuicultura
(Universidad Santo Tomás, 2026-08-26) Cerón Lombana, Juan Carlos; Núñez Rodríguez, Rafael Augusto; Peña Gálviz, Omar Leonardo; Universidad santo Tomás
Intensive tilapia aquaculture faces toxicity risks from nitrogenous compounds, whose nonlinear behavior makes them difficult to predict with conventional monitoring. To bridge the gap between water‑quality diagnosis and on‑farm action, this study validates a neural soft sensor and a prescriptive Decision Support System (DSS) as an integrated architecture for nitrogen‑risk management. Methodologically, a dataset of 291.812 records was consolidated by combining open observational sources with synthetic scenarios of critical toxicity. Five machine‑learning architectures were evaluated on this corpus, and a deterministic DSS based on biological rules for Oreochromis spp. was designed. The results show that a multilayer perceptron (MLP) achieved a macro‑F1 of 0.9964 and a mean latency of 0.53 ms, an ideal range for IoT deployments. These metrics are interpreted as a validation of computational feasibility under a labelling scheme derived from deterministic rules, rather than as a guarantee of immediate generalization to field conditions. When integrated into the DSS, this neural core achieved a zero under‑prescription rate in high‑risk scenarios and a conservative over‑prescription rate of about 60%, supporting a safety‑first design policy that avoids omitting critical interventions. The prototype provides a solid methodological reference for future intelligent systems in regional aquaculture.
Diseño de un modelo metodológico para el fortalecimiento de la gestión de indicadores institucionales mediante analítica de datos en la Alcaldía de Bucaramanga
(Universidad Santo Tomás, 2026-07-10) González Peralta, Manuel Antonio; Escobar Mejía, Marycruz; Páez Casas, Deisy Carolina; Universidad Santo Tomás
The Bucaramanga Mayor's Office lacked a consolidated and standardized inventory of its institutional performance indicators, resulting in methodological fragmentation across government departments and non-compliance with the standards established by the National Statistical System (NSS), thereby limiting evidence-based decision-making. This study aimed to design a methodological model to strengthen the management of institutional performance indicators through data analytics by integrating a technical diagnostic assessment, a standardized documentation instrument, and a digital visualization tool. A mixed-methods business internship was conducted over 480 hours and structured into three phases: (1) diagnosis of 46 indicators from eight departments against the 21 mandatory NSS fields using an Extract, Transform, and Load (ETL) process implemented in Python; (2) redesign of the official Indicator Data Sheet and development of an institutional strengthening plan; and (3) development of a Power BI dashboard prototype. The results showed a weighted average completeness of 37.7% across the assessed indicators, a new NSS-aligned Indicator Data Sheet, an institutional strengthening plan comprising nine strategies distributed across three implementation horizons, and a functional dashboard deployed within the institutional environment containing 86 indicators from nine departments. The findings demonstrate that transforming indicator management in public institutions depends not only on the technical quality of management instruments but also on effective organizational change management, particularly during periods of administrative transition.
Identificación paramétrica de rigidez y amortiguamiento estructural mediante una Red Neuronal Evolutiva con optimización multiobjetivo
(Universidad Santo Tomás, 2026-06-26) Niño Sandoval, Kelly Johanna; Valencia Niño, Cesar Hernando; Rodríguez Torres, Andrés José; Universidad Santo Tomás
Problem: the identification of structural parameters in multiple-degree-of-freedom systems using only acceleration data constitutes a coupled inverse problem, sensitive to noise and characterized by limited damping observability, in this context, objective: this research developed a parametric identification framework to estimate the stiffness 𝑘𝑖 and damping 𝑐𝑖 of the five stories of a reducedscale prototype building, relying exclusively on accelerometric signals, to this end, method: a physics-informed Evolutionary Artificial Neural Network was integrated with a two-phase NSGAII Multiobjective Genetic Algorithm, formulating the identification task as the simultaneous minimization of temporal fidelity, modal consistency, and physical regularization; the network mapped a feature vector of dimension D=660 to admissible structural parameters through bounded decoding. As results: using the 1985 Mexico City earthquake record, nine configurations were evaluated with 30 independent runs, and the best architecture (H1=32, H2=16, PS1=220) achieved a median RMSE of 0.1895 m/s², a median 𝑅 2 of 0.860, and coefficients of variation below 0.15 for both stiffness and damping; additionally, nonparametric validation showed significant differences compared with GA and PSO (p<0.001), and the Pareto front reached 98 solutions with HV=1.000±pm0.000; finally, discussion: the results indicate that the proposed framework reduces repeated identification to a single learning task, with millisecond-level inference and potential application in post-earthquake structural health monitoring.
Clasificación del Desempeño Académico en la Prueba Saber 11 Mediante Variables Socioeconómicas: Estudio Comparativo Entre Redes Neuronales, Convolucionales y Sistemas Expertos
(Universidad Santo Tomás, 2026-06-25) Sanabria Casanova, Cesar Augusto; Acosta Velásquez, Elkin Vladimir; Valencia Niño, Cesar Hernando; Universidad Santo Tomas
The research work was developed as a comparative study of different artificial intelligence models ("MLP – CNN 1D - LightGBM") for the classification of academic performance based on the results of the Saber 11° State examination in Colombia, through socioeconomic, family, and institutional variables. It addresses the problem of the limitations in traditional statistical approaches when capturing the complex, non-linear, and high-dimensional structures of the variables that explain educational gaps in Colombia. To this end, multiclass predictive models were formulated to utilize a historical dataset of approximately 7.2 million anonymized records provided by the ICFES, corresponding to the periods from 2014-2 to 2024-2. The CRISP-DM (Cross-Industry Standard Process for Data Mining) methodology was employed, implementing a hybrid processing pipeline and comparative analyses of the three classes of artificial intelligence models. The experimental results demonstrated that deep neural network architectures ("MLP") achieved the best predictive performance and highest statistical stability, outperforming the one-dimensional convolutional neural network ("1D CNN") and the ensemble algorithm (LightGBM) by capturing the non-linear dependencies of the dataset for each individual. In conclusion, socioeconomic variables possess a significant predictive capacity regarding academic performance classification, and deep learning models serve as a technological and methodological tool to validate the design of an early warning system and support evidence-based decision-making in the Colombian educational sector.

Examinar

Envíos recientes