Evaluación de la calidad de los datos de mobile crowdsensing mediante técnicas de inteligencia artificial.

Abstract

Abstract: Mobile Crowdsensing (MCS) has become established as an alternative for large-scale environmental monitoring; however, sensor heterogeneity and spatial and temporal variability complicate its evaluation. This thesis proposes an unsupervised approach to assess the quality of MCS data from geographically distributed fixed sensors, developed using the CRISP-DM methodology and considering dimensions such as consistency, completeness, accuracy, and anomaly detection. The proposal is based on two independent models. The first, focused on spatial analysis, integrates PCA with K-Means and HDBSCAN to identify geographic structures and heterogeneous clusters. The results show multiple clusters and a significant percentage of noise, demonstrating sensitivity to irregular spatial distributions. The second model, LSTM-AE + Isolation Forest, focused on temporal analysis, exhibited greater stability and coherence with consistency metrics. In each evaluation, a consensus rule was applied that integrates statistical, spatial, and temporal criteria. The system was implemented using Docker, Angular 17, and Python APIs. It is concluded that validated MCS data constitute a reliable contribution to decision-making in various fields.