Análise estatística utilizando modelos de machine learning em redes móveis
Resumo
Resumo: Este trabalho investiga a relação entre indicadores de desempenho de redes móveis (KPIs) e a Qualidade de Experiência (QoE) dos usuários, medida por Crowdtesting. Para superar diferenças de granularidade e ausência de identificação de células nos testes, os dados foram equalizados por clusters de latitude/longitude e agregados em nível diário. Modelos de regressão penalizada (LASSO, Ridge e Elastic Net) foram aplicados utilizando a linguagem R e o pacote glmnet. Os resultados indicaram que o modelo Ridge apresentou a melhor área sob a curva ROC (AUC = 0,625), enquanto o Elastic Net atingiu o melhor equilíbrio entre acurácia e parcimônia (AUC = 0,5536 com 58 variáveis). Conclui-se que a integração entre KPIs e Crowdtesting, mediada por técnicas de machine learning, oferece uma abordagem possível para predição da QoE em redes móveis Abstract: This work investigates the relationship between mobile network performance indicators (KPIs) and user Quality of Experience (QoE), measured through Crowdtesting. To address differences in granularity and lack of cell identification, the data were equalized by latitude/longitude clusters and aggregated at a daily level. Penalized regression models (LASSO, Ridge, and Elastic Net) were applied using the R language and the glmnet package. Results indicated that Ridge achieved the best area under the ROC curve (AUC = 0.625), while Elastic Net provided the best balance between accuracy and parsimony (AUC = 0.5536 with 58 variables). It is concluded that integrating KPIs and Crowdtesting, mediated by machine learning techniques, offers a possible approach to predicting QoE in mobile networks
Collections
- Data Science & Big Data [163]