Influencia del pre-procesamiento de datos dentro del desempeño de modelos de perfilamiento de clientes elaborados con herramientas de minería de datos

摘要

In this research we aim to understand, associate and quantitatively identify the performance of customer profiling models with data pre-processing techniques, as well as specify best practices for the pre-processing of data related with the solution of customer profiling problems with data mining tools. We present the three different stages of our work, namely: 1. Construction of the case study and data collection in which a dataset of the Data Mining Cup is selected ; 2. Generation of experimental data, in which the pre-processing is executed through three methods, multicluster feature selection (MCFS), Laplacian score - LPSC and spectral feature selection (SPEC), and grouping using K-mean techniques, hierarchical grouping and self-organizing maps; 3. Comparison and analysis of results describing the best practices for the pre-processing, resulting from the experiments.