1 February 2017 in Knowledge-Based Systems
20 August 2019 in IEEE Transactions on Fuzzy Systems
One of the best-known and most effective methods in supervised classification is the k nearest neighbors algorithm (kNN). Several approaches have been proposed to improve its accuracy, where fuzzy approaches prove to be among the most successful, highlighting the classical Fuzzy k-nearest neighbors (FkNN). However, these traditional algorithms fail to tackle the large amounts of data that are available today. There are multiple alternatives to enable kNN classification in big datasets, spotlighting the approximate version of kNN called Hybrid Spill Tree. Nevertheless, the existing proposals of FkNN for big data problems are not fully scalable, because a high computational load is required to obtain the same behavior as the original FkNN algorithm. This work proposes Global Approximate Hybrid Spill Tree FkNN and Local Hybrid Spill Tree FkNN, two approximate approaches that speed up runtime without losing quality in the classification process. The experimentation compares various FkNN approaches for big data with datasets of up to 11 million instances. The results show an improvement in runtime and accuracy over literature algorithms.
1 January 2020 in IEEE Access
It is recognized the importance of knowing the descriptive properties of a dataset when tackling a data science problem. Having information about the redundancy, complexity and density of a problem allows us to make decisions as to which data preprocessing and machine learning techniques are most suitable. In classification problems, there are multiple metrics to describe the overlapping of the features between classes, class imbalances or separability, among others. However, these metrics may not scale up well when dealing with big datasets, or may not simply be sufficiently informative in this context. In this paper, we provide a package of metrics for big data classification problems. In particular, we propose two new big data metrics: Neighborhood Density and Decision Tree Progression, which study density and accuracy progression by discarding half of the samples. In addition, we enable a number of basic metrics to handle big data. The experimental study carried out in standard big data classification problems shows that our metrics can quickly characterize big datasets. We identified a clear redundancy of information in most datasets, so that, discarding randomly 75% of the samples does not drastically affect the accuracy of the classifiers used. Thus, the proposed big data metrics, which are available as a Spark-Package, provide a fast assessment of the shape of a classification dataset prior to applying big data preprocessing, toward smart data.