Thesis Sistema de recomendación y estandarización de roles SAP integrando inteligencia artificial en la gestión de accesos para entornos empresariales
Loading...
Date
2026-03
Authors
Journal Title
Journal ISSN
Volume Title
Program
Ingeniería Civil Telemática
Departament
Campus
Campus Santiago San Joaquín
Abstract
Se diseñó, implementó y evaluó un sistema web de recomendación y estandarización de roles SAP mediante aprendizaje automático, orientado a reducir permisos durmientes en entornos empresariales. El sistema compara usuarios del mismo cargo y sucursal usando similitud de Jaccard para identificar roles faltantes, y los valida con un clasificador CatBoost con umbral ω = 0,7, configuración que prioriza la precisión sobre el recall en línea con el principio de menor privilegio. La retroalimentación del equipo de gestión de accesos queda registrada para un posterior proceso de aprendizaje asistido, en el que un administrador puede incorporar manualmente las etiquetas validadas al conjunto de entrenamiento. La evaluación del sistema integró tres dimensiones complementarias: el desempeño en tareas —medido mediante tiempos de completitud y errores observados—, la satisfacción percibida —recogida con los instrumentos ASQ y SUS— y los hallazgos cualitativos obtenidos en entrevistas semiestructuradas. La triangulación de estas dimensiones con tres participantes —dos expertos en gestión IAM y un usuario general— arrojó un puntaje SUS promedio de 76,7 sobre 100 (categoría aceptable) y un promedio ASQ de 6,0 sobre 7, con todas las tareas completadas sin asistencia. En cuanto al impacto operativo, el sistema reduce el tiempo de procesamiento desde 20 minutos por usuario en el proceso manual a 0,005 minutos al procesar lotes de 1.000 trabajadores, lo que representa una reducción del 99,97%, liberando al equipo de gestión de una carga operativa significativa.
A web-based SAP role recommendation and standardization system was designed, implemented and evaluated using machine learning techniques, aimed at reducing dormant permissions in enterprise environments. The system compares users sharing the same job title and branch via Jaccard similarity to identify missing roles, and validates candidates using a CatBoost classifier with threshold ω = 0,7, a configuration that prioritizes precision over recall in accordance with the principle of least privilege. Feedback from the access management team is recorded to support a future assisted learning process, in which an administrator can manually incorporate validated labels into the training set. The system evaluation integrated three complementary dimensions: task performance —measured through completion times and observed errors—, perceived satisfaction —captured with the ASQ and SUS instruments— and qualitative findings from semi-s--tructured interviews. The triangulation of these dimensions across three participants —two IAM management experts and one non-specialized user— yielded an average SUS score of 76,7 out of 100 (Acceptable category) and an average ASQ score of 6,0 out of 7, with all tasks completed without assistance. Regarding operational impact, the system reduces processing time from 20 minutes per user in the manual process to 0,005 minutes when processing batches of 1,000 workers, representing a 99,97 % reduction and freeing the access management team from significant operational overhead.
A web-based SAP role recommendation and standardization system was designed, implemented and evaluated using machine learning techniques, aimed at reducing dormant permissions in enterprise environments. The system compares users sharing the same job title and branch via Jaccard similarity to identify missing roles, and validates candidates using a CatBoost classifier with threshold ω = 0,7, a configuration that prioritizes precision over recall in accordance with the principle of least privilege. Feedback from the access management team is recorded to support a future assisted learning process, in which an administrator can manually incorporate validated labels into the training set. The system evaluation integrated three complementary dimensions: task performance —measured through completion times and observed errors—, perceived satisfaction —captured with the ASQ and SUS instruments— and qualitative findings from semi-s--tructured interviews. The triangulation of these dimensions across three participants —two IAM management experts and one non-specialized user— yielded an average SUS score of 76,7 out of 100 (Acceptable category) and an average ASQ score of 6,0 out of 7, with all tasks completed without assistance. Regarding operational impact, the system reduces processing time from 20 minutes per user in the manual process to 0,005 minutes when processing batches of 1,000 workers, representing a 99,97 % reduction and freeing the access management team from significant operational overhead.
Description
Keywords
Control de acceso basado en roles(RBAC), Permisos durmientes, Usabilidad, Métricas de proceso, Role-based access control(RBAC), Dormant permissions, Usability, Process metrics
