Intelligent Data Analysis: An International Journal

Modern data collections create vast opportunities for detecting useful hidden relationships. Also, increasingly, they fuel data privacy concerns. A trade-off between privacy protection and data usefulness is by now widely acknowledged. Real world data classification tasks, as for example credit scoring applications have to deal with such data security limitations by finding a way to effectively incorporate privacy preserving procedures. To this end we propose as a first stage to use a microaggregation procedure in order to anonymize data over personal credit client feature information. In a second stage we examine the performance of support vector machines (SVM) on such anonymized data. SVM are powerful and robust machine learning methods, having superior credit scoring classification performance when applied to original, non-anonymized data. We first partition the original credit scoring data set and construct anonymized data representatives, which are then used for credit client behavior forecasting models constructed by SVM and other comparable learning methods. The validation procedure for such models is adapted to the two-stage modeling approach. In order to assess the loss owing to data anonymization, the different classification models are evaluated against models that are trained on the original data.

Attempts to mine text documents to discover deviations or anomalies have increased in recent years due to the elevated amount of textual data in today's data repositories. Text mining assists in uncovering hidden information contents across multiple documents. Although various text mining tools are available, their focus is mainly to assist in data summarization or document classification. These tasks proved to be helpful, however; they do not provide semantic analysis and rigorous textual comparison to detect abnormal sentences that exist in the documents. In this paper, we describe a text mining system that is able to detect sentence deviations from a collection of financial documents. The system implements a dissimilarity function to compare sentences represented as graphs. Our evaluation on the proposed system revolves around experiments using financial statements of a bank. The findings provide valid evidence that the proposed system is able to identify deviating sentences occurring in the documents. The detected deviations can be beneficial for the authorities in order to improve their business decisions.

In recent years, data reduction techniques have gained scene in applied economics. Following this trail, we apply and extend the use of a data reduction technique such as principal component analysis (PCA), to assess the degree of systemic risk in the Chilean banking system. In particular, we address the following questions: (i) to what extent the degree of common risk exposure in the banking system has changed over the past decades; (ii) during which periods this exposure increased the most; and (iii) based on predefined thresholds, when this commonality has become systemically relevant. Additionally, we identify systemically important financial institutions (SIFIs) based on their contribution to the degree of common risk exposure during periods of higher systemic risk. We find that prior to the 2008-09 global financial crisis the degree of common risk exposure in Chile increased significantly, and that the banks that contributed the most were not necessarily the biggest ones in size, as measured by their assets share.

We examine the task of finding thematic structure in a data corpus comprising text and time series. To achieve this we introduce topic factor modelling (TFM). We develop a novel, joint generative model for both data types which resembles supervised latent Dirichlet allocation. TFM allows the decomposition of time series into factors which also reflect the thematic content of the text. We describe a variational method for inference and demonstrate its effectiveness on a synthetic corpus. For a corpus of publicly available equity data, we show that a TFM can simultaneously and robustly model both stock price time series and text data describing the corresponding companies. We also discuss how topic modelling could assist with external tasks such as robust covariance estimation.

Credit scoring is an automated, objective and consistent tool which helps lenders to provide quick loan decisions. It can replace some of the more mechanical work done by experienced loan officers whose decisions are intuitive but potentially subject to bias. Prospective borrowers may have a strong motivation to fraudulently falsify one or more of the attributes they report on their application form. Applicants learn about the characteristics that are used to build credit scoring models, and may alter the answers on their application form to improve their chance of loan approval. Few automated credit scoring models have considered falsified information from borrowers. We will show that sometimes it is profitable for financial institutions to spend money and effort to identify dishonest customers. We will also find the optimal effort that banks should spend on identifying these liars. Furthermore, we will show that it is possible for liars to eventually adjust their lies to escape from credit checks. The proposed issue will be studied using simulated data and discriminant analysis. This research can help lending financial institutions to reduce risk and maximize profit, and it also shows that it is feasible for customers to lie intelligently so as to evade credit checks and get loans.

Clustering groups objects based on their similarity using unsupervised learning. Clustering is an NP hard problem. A number of clustering algorithms use heuristics to create a reasonable grouping of objects. However, clustering schemes created by different heuristic algorithms do not always completely agree with each other. For example, an object may belong to different clusters for different algorithms. Therefore, researchers have proposed a number of clustering ensemble techniques to combine the clustering schemes from different algorithms. This paper proposes a Rough Set based ensemble method for preserving the inherent order in clustering. The proposal is demonstrated with the help of daily price patterns of commodities, which are grouped based on Black Scholes volatility index as well as the distribution of prices.

This work addresses the issue of high dimensionality for linear multiclass Support Vector Machines (SVMs) using second-order cone programming (SOCP) formulations. These formulations provide a robust and efficient framework for classification, while an adequate feature selection process may improve predictive performance. We extend the ideas of SOCP-SVM from binary to multiclass classification, while a sequential backward elimination algorithm is proposed for variable selection, defining a contribution measure to determine the feature relevance. Experimental results with multiclass microarray datasets demonstrate the effectiveness of a low-dimensional data representation in terms of performance.

An empirical framework for customer churn prediction modeling is presented in this work. This task represents a very interesting business analytics challenge, given its highly class imbalanced nature, and the presence of noisy variables that adversely affect the prediction capabilities of classification models. In this work, two SVM-based techniques are compared: Support Vector Data Description (SVDD), and standard two-class SVMs. The proposed methodology involves the comparison of these two methods under different conditions of class imbalance and using different subsets of variables. Feature ranking is performed via the Fisher Score Criterion, while the class imbalance problem is dealt with through resampling techniques, namely random undersampling and SMOTE oversampling. Experiments on four customer churn prediction datasets show the advantages of SVDD: it outperforms standard SVM in terms of predictive performance, demonstrating the importance of techniques that take the class imbalance problem into account.

Data Mining (DM) researchers often focus on the development and testing of models for a single decision (e.g., direct mailing, churn detection, etc.). In practice, however, multiple decisions have often to be made simultaneously which are not independent and the best global solution is often not the combination of the best individual solutions. This problem can be addressed by searching for the overall best solution by using optimization methods based on the predictions made by the DM models. We describe one case study were this approach was used to optimize the layout of a retail store in order to maximize predicted sales. A metaheuristic is used to search different hypothesis of space allocations for multiple product categories, guided by the predictions made by regression models that estimate the sales for each category based on the assigned space. We test three metaheuristics and three regression algorithms on this task. Results show that the Particle Swam Optimization method guided by the models obtained with Random Forests and Support Vector Machines models obtain good results. We also provide insights about the relationship between the correctness of the regression models and the metaheuristics performance.

The share of the services offered via the Internet by nowadays banking companies is quickly growing, making of the understanding of online customers one of the major concerns. Data mining tools have proven their efficiency in addressing this challenge by providing unsupervised quantitative techniques to identify those segments of customers with similar characteristics. This paper will focus on segmenting an online banking customer base in a meaningful way for the business by enhancing an unsupervised quantitative technique approach with domain knowledge. Both traditional and knowledge-based approaches will be applied and evaluated. Thanks to an extensive description and discussion of the new insights, the complementarity of the two approaches is illustrated.