The article proposes an expert system for detection, and subsequent investigation, of groups of collaborating automobile insurance fraudsters. The system is described and examined in great detail, several technical difficulties in detecting fraud are also considered, for it to be applicable in practice. Opposed to many other approaches, the system uses networks for representation of data. Networks are the most natural representation of such a relational domain, allowing formulation and analysis of complex relations between entities. Fraudulent entities are found by employing a novel assessment algorithm, Iterative Assessment Algorithm (IAA), also presented in the article. Besides intrinsic attributes of entities, the algorithm explores also the relations between entities. The prototype was evaluated and rigorously analyzed on real world data. Results show that automobile insurance fraud can be efficiently detected with the proposed system and that appropriate data representation is vital. Key words: Fraud detection, Automobile insurance, Social network analysis, Link analysis, Assessment propagation 1. Introduction Fraud is encountered in a variety of domains. It comes in all different shapes and sizes, from traditional fraud, e.g. Such groups can be found in the automobile insurance domain. Here fraudsters stage traffic accidents and issue fake insurance claims to gain (unjustified) funds from their general or vehicle insurance. There are also cases where an accident has never occurred, and the vehicles have only been placed onto the road. Still, the majority of such fraud is not planned (opportunistic fraud) - an individual only seizes the opportunity arising from the accident and issues exaggerated insurance claims or claims for past damages. Staged accidents have several common characteristics. They occur in late hours and non-urban areas in order to reduce the probability of witnesses. Drivers are usually younger males, there are many passengers in the vehicles, but never children or elders. The police is always called to the scene to make the subsequent acquisition of means easier. It is also not uncommon that all of the participants have multiple (serious) injuries, when there is almost no damage on the vehicles. Many other suspicious characteristics exist, not mentioned here.

The goal of our research is to find patterns of EEG activity that will allow us to correctly identify seizures in living rats using machine learning techniques. Features are extracted from the EEG to characterize the signal over time. We perform model selection to reduce the set of features, as the goal is to have the algorithm running on a small personal device. The chosen features are used within a supervised classifier, based on randomized forests, in order to separate the different brain states. One of the challenges of this research is to detect all seizures, while preserving a low false positive rate, and low detection latency. We present results showing we can achieve this using data from three separate animals. The long-term goal of this research is to use this seizure detection method as part of a closed-loop adaptive neuro-stimulation device to reduce the incidence and duration of seizures.

The ensemble approach for statistical modelling was first made popular by such algorithms as boosting (Freund and Schapire 1996; Friedman et al. 2000), bagging (Breiman 1996), random forest (Breiman 2001), and the gradient boosting machine (Friedman 2001). They are powerful algorithms for solving prediction problems. This article is concerned with using the ensemble approach for a different problem, variable selection. We shall use the terms "prediction ensemble" and "variableselection ensemble" to differentiate ensembles used for these different purposes.

A computational challenge to validate the candidate disease genes identified in a high-throughput genomic study is to elucidate the associations between the set of candidate genes and disease phenotypes. The conventional gene set enrichment analysis often fails to reveal associations between disease phenotypes and the gene sets with a short list of poorly annotated genes, because the existing annotations of disease causative genes are incomplete. We propose a network-based computational approach called rcNet to discover the associations between gene sets and disease phenotypes. Assuming coherent associations between the genes ranked by their relevance to the query gene set, and the disease phenotypes ranked by their relevance to the hidden target disease phenotypes of the query gene set, we formulate a learning framework maximizing the rank coherence with respect to the known disease phenotype-gene associations. An efficient algorithm coupling ridge regression with label propagation, and two variants are introduced to find the optimal solution of the framework. We evaluated the rcNet algorithms and existing baseline methods with both leave-one-out cross-validation and a task of predicting recently discovered disease-gene associations in OMIM. The experiments demonstrated that the rcNet algorithms achieved the best overall rankings compared to the baselines. To further validate the reproducibility of the performance, we applied the algorithms to identify the target diseases of novel candidate disease genes obtained from recent studies of GWAS, DNA copy number variation analysis, and gene expression profiling. The algorithms ranked the target disease of the candidate genes at the top of the rank list in many cases across all the three case studies. The rcNet algorithms are available as a webtool for disease and gene set association analysis at http://compbio.cs.umn.edu/dgsa_rcNet.

Privacy-preserving machine learning algorithms are crucial for the increasingly common setting in which personal data, such as medical or financial records, are analyzed. We provide general techniques to produce privacy-preserving approximations of classifiers learned via (regularized) empirical risk minimization (ERM). These algorithms are private under the $\epsilon$-differential privacy definition due to Dwork et al. (2006). First we apply the output perturbation ideas of Dwork et al. (2006), to ERM classification. Then we propose a new method, objective perturbation, for privacy-preserving machine learning algorithm design. This method entails perturbing the objective function before optimizing over classifiers. If the loss and regularizer satisfy certain convexity and differentiability criteria, we prove theoretical results showing that our algorithms preserve privacy, and provide generalization bounds for linear and nonlinear kernels. We further present a privacy-preserving technique for tuning the parameters in general machine learning algorithms, thereby providing end-to-end privacy guarantees for the training process. We apply these results to produce privacy-preserving analogues of regularized logistic regression and support vector machines. We obtain encouraging results from evaluating their performance on real demographic and benchmark data sets. Our results show that both theoretically and empirically, objective perturbation is superior to the previous state-of-the-art, output perturbation, in managing the inherent tradeoff between privacy and learning performance.

Inspired by biological vision systems, the over-complete local features with huge cardinality are increasingly used for face recognition during the last decades. Accordingly, feature selection has become more and more important and plays a critical role for face data description and recognition. In this paper, we propose a trainable feature selection algorithm based on the regularized frame for face recognition. By enforcing a sparsity penalty term on the minimum squared error (MSE) criterion, we cast the feature selection problem into a combinatorial sparse approximation problem, which can be solved by greedy methods or convex relaxation methods. Moreover, based on the same frame, we propose a sparse Ho-Kashyap (HK) procedure to obtain simultaneously the optimal sparse solution and the corresponding margin vector of the MSE criterion. The proposed methods are used for selecting the most informative Gabor features of face images for recognition and the experimental results on benchmark face databases demonstrate the effectiveness of the proposed methods.

Unsupervised two-view learning, or detection of dependencies between two paired data sets, is typically done by some variant of canonical correlation analysis (CCA). CCA searches for a linear projection for each view, such that the correlations between the projections are maximized. The solution is invariant to any linear transformation of either or both of the views; for tasks with small sample size such flexibility implies overfitting, which is even worse for more flexible nonparametric or kernel-based dependency discovery methods. We develop variants which reduce the degrees of freedom by assuming constraints on similarity of the projections in the two views. A particular example is provided by a cancer gene discovery application where chromosomal distance affects the dependencies between gene copy number and activity levels. Similarity constraints are shown to improve detection performance of known cancer genes.

A variable screening procedure via correlation learning was proposed Fan and Lv (2008) to reduce dimensionality in sparse ultra-high dimensional models. Even when the true model is linear, the marginal regression can be highly nonlinear. To address this issue, we further extend the correlation learning to marginal nonparametric learning. Our nonparametric independence screening is called NIS, a specific member of the sure independence screening. Several closely related variable screening procedures are proposed. Under the nonparametric additive models, it is shown that under some mild technical conditions, the proposed independence screening methods enjoy a sure screening property. The extent to which the dimensionality can be reduced by independence screening is also explicitly quantified. As a methodological extension, an iterative nonparametric independence screening (INIS) is also proposed to enhance the finite sample performance for fitting sparse additive models. The simulation results and a real data analysis demonstrate that the proposed procedure works well with moderate sample size and large dimension and performs better than competing methods.

Performing effective preference-based data retrieval requires detailed and preferentially meaningful structurized information about the current user as well as the items under consideration. A common problem is that representations of items often only consist of mere technical attributes, which do not resemble human perception. This is particularly true for integral items such as movies or songs. It is often claimed that meaningful item features could be extracted from collaborative rating data, which is becoming available through social networking services. However, there is only anecdotal evidence supporting this claim; but if it is true, the extracted information could very valuable for preference-based data retrieval. In this paper, we propose a methodology to systematically check this common claim. We performed a preliminary investigation on a large collection of movie ratings and present initial evidence.

When choosing a suitable technique for regression and classification with multivariate predictor variables, one is often faced with a tradeoff between interpretability and high predictive accuracy. To give a classical example, classification and regression trees are easy to understand and interpret. Tree ensembles like Random Forests provide usually more accurate predictions. Yet tree ensembles are also more difficult to analyze than single trees and are often criticized, perhaps unfairly, as `black box' predictors. Node harvest is trying to reconcile the two aims of interpretability and predictive accuracy by combining positive aspects of trees and tree ensembles. Results are very sparse and interpretable and predictive accuracy is extremely competitive, especially for low signal-to-noise data. The procedure is simple: an initial set of a few thousand nodes is generated randomly. If a new observation falls into just a single node, its prediction is the mean response of all training observation within this node, identical to a tree-like prediction. A new observation falls typically into several nodes and its prediction is then the weighted average of the mean responses across all these nodes. The only role of node harvest is to `pick' the right nodes from the initial large ensemble of nodes by choosing node weights, which amounts in the proposed algorithm to a quadratic programming problem with linear inequality constraints. The solution is sparse in the sense that only very few nodes are selected with a nonzero weight. This sparsity is not explicitly enforced. Maybe surprisingly, it is not necessary to select a tuning parameter for optimal predictive accuracy. Node harvest can handle mixed data and missing values and is shown to be simple to interpret and competitive in predictive accuracy on a variety of data sets.