FDR2-BD: A Fast Data Reduction Recommendation Tool for Tabular Big Data Classification Problems

In this paper, a methodological data condensation approach for reducing tabular big datasets in classification problems is presented, named FDR2-BD. The key of our proposal to analyze dual way (vertical and horizontal), so as provide smart combination between feature selection generate dense clusters uniform sampling reduction keep only few representative samples from each problem area. Its main advantage allowing the model’s predictive quality be kept range determined by user’s threshold. robustness built on hyper-parametrization process, which all are taken into consideration following k-fold procedure. Another significant capability being fast scalable using fully optimized parallel operations provided Apache Spark. An extensive experimental study performed over 25 with different characteristics. most cases, obtained percentages above 95%, thus outperforming state-of-the-art solutions such FCNN_MR that barely reach 70%. promising outcome maintaining representativeness original information, prediction values around 1% baseline.