Resumen

Supervised Algorithms for Failure Prediction in the UPSE LAN network. RCTU [online]. 2025, vol.12, n.2, pp.90-107. ISSN 1390-7697.  https://doi.org/https://doi.org/10.26423/03cdzf27.

In university networks such as the one at the Universidad Estatal Península de Santa Elena, LAN management remains predominantly reactive, lacking historical records and early-warning mechanisms. This study proposes the application of supervised machine-learning algorithms, selected based on scientific evidence, to build and evaluate a predictive failure-detection model using SNMP telemetry collected through Zabbix. A combined Design Science Research (DSR) and CRISP-DM methodology was applied, with 60-minute windows over 7 571 samples (729 failures and 6 842 normal cases). Two approaches were compared: a Random Forest model trained on statistical features, and a one-dimensional convolutional neural network applied to multivariate sequences. Random Forest achieved an accuracy of 96.88 %, while the neural network reached a recall of 73,10 %. The results show the complementary nature of both models and demonstrate that their combined use supports proactive institutional network management, reducing response times to incidents.

Palabras clave : Machine learning; Predictive analytics; Network management; University networks; SNMP.