The Power of Sampling and Stacking for the PaKDD-2007 Cross-Selling Problem

Abstract

This article presents an efficient solution for the PAKDD-2007 Competition cross-selling problem. The solution is based on a thorough approach which involves the creation of new input variables, efficient data preparation and transformation, adequate data sampling strategy and a combination of two of the most robust modeling techniques. Due to the complexity imposed by the very small amount of examples in the target class, the approach for model robustness was to produce the median score of the 11 models developed with an adapted version of the 11-fold cross-validation process and the use of a combination of two robust techniques via stacking, the MLP neural network and the n-tuple classifier. Despite the problem complexity, the performance on the prediction data set (unlabeled samples), measured through KS2 and ROC curves was shown to be very effective and finished as the first runner-up solution of the competition.

The role of temporal feature extraction and bagging of MLP neural networks for solving the WCCI 2008 Ford Classification Challenge

“…She concluded that MLP neural networks were the most robust against performance degradation. More recently, several applications have confirmed the robustness of these neural networks on real-world problems [8,9].…”

“…It has also been successfully used in data mining applications either isolate [17] or in ensembles [8,9]. However, one drawback of this technique is the need of a validation (holdout) data set for preventing over fitting, when there are not much labeled data available.…”

“…This mixture of techniques has been successfully applied to complex problems in different actual domains [8,9]. Feature extraction embeds the domain experts' knowledge into the data via random variables which map the original space into a feature space much more meaningful to the problem being tackled.…”

Data Mining Solution for Assessing Brazilian Secondary School Quality Based on ENEM and Census Data

“…Therefore, this paper will use the area under the ROC curve (AUC_ROC) [Provost and Fawcett, 2001], one the most widely accepted performance metrics for binary classifiers. This paper will also use the maximum Kolmogorov-Smirnov distance (Max_KS2) as a dissimilarity metrics [Conover, 1999] to assess continuous score classifier output, as commonly done in financial applications [Adeodato et al, 2008].…”

pRAM n-tuple Classifier - a new architecture of probabilistic RAM neurons for classification problems

“…Three years ago, Adeodato et al [12] have successfully applied a modified version of the n-tuple classifier to a real problem in the PAKDD 2007 data mining competition. That system has evolved and this paper presents the updated version ofthis approach which combines the features of: 1. an architecture similar to the n-tuple classifier, 2. the pRAM neuron with a slightly different recall mode and 3. additive Gaussian noise training, without any constraints for classification problems.…”