Proceedings of the 2010 Conference on Intelligent Data Understanding

New Frontiers in Applied Data Mining

Knowledge Discovery from Sensor Data
nash.edu.au Wide-area sensor infrastructures, remote sensors, RFIDs, and wireless sensor networks yield massive volumes of disparate, dynamic, and geographically distributed data. As such sensors are becoming ubiquitous, a set of broad requirements is beginning to emerge across high-priority applications including adaptability to climate change, electric grid monitoring, disaster preparedness and management, national or homeland security, and the management of critical infrastructures. The raw data from sensors need to be efficiently managed and transformed to usable information through data fusion, which in turn must be converted to predictive insights via knowledge discovery, ultimately facilitating automated or human-induced tactical decisions or strategic policy based on decision sciences and decision support systems. Keeping in view the requirements of the emerging field of knowledge discovery from sensor data, we took initiative to develop a community of researchers with common interests and scientific goals, which culminated into the organization of SensorKDD series of workshops in conjunction with the
Complex Networks In Climate Science Progress%2C Opportunities And Challenges.

New perspectives and methods in link prediction.

Generating Diverse Ensembles to Counter the Problem of Class Imbalance.

User Generated Content Consumption and Social Networking in Knowledge%2DSharing OSNs.

A Robust Decision Tree Algorithm for Imbalanced Data Sets.

A Private and Reliable Recommendation System for Social Networks.

Data Mining for Imbalanced Datasets An Overview.

Time to CARE a collaborative engine for practical disease prediction.

A statistical approach to finding overlooked genetic associations.

Time to CARE a collaborative engine for practical disease prediction.

Identifying and evaluating community structure in complex networks.

An exploration of climate data using complex networks.
To discover patterns in historical data, climate scientists have applied various clustering methods with the goal of identifying regions that share some common climatological behavior. However, past approaches are limited by the fact that they either consider only a single time period (snapshot) of multivariate data, or they consider only a single variable by using the time series data as multi-dimensional feature vector. In both cases, potentially useful information may be lost. Moreover, clusters in high-dimensional data space can be difficult to interpret, prompting the need for a more effective data representation. We address both of these issues by employing a complex network (graph) to represent climate data, a more intuitive model that can be used for analysis while also having a direct mapping to the physical world for interpretation. A cross correlation function is used to weight network edges, thus respecting the temporal nature of the data, and a community detection algorithm identifies multivariate clusters. Examining networks for consecutive periods allows us to study structural changes over time. We show that communities have a climatological interpretation and that disturbances in structure can be an indicator of climate events (or lack thereof). Finally, we discuss how this model can be applied for the discovery of more complex concepts such as unknown teleconnections or the development of multivariate climate indices and predictive insights.
Proceedings of the Third International Workshop on Knowledge Discovery from Sensor Data

Modeling a Store%27s Product Space as a Social Network.

Applying Learning Algorithms to Music Generation.

An exploration of climate data using complex networks.
To discover patterns in historical data, climate scientists have applied various clustering methods with the goal of identifying regions that share some common climatological behavior. However, past approaches are limited by the fact that they either consider only a single time period (snapshot) of multivariate data, or they consider only a single variable by using the time series data as multi-dimensional feature vector. In both cases, potentially useful information may be lost. Moreover, clusters in high-dimensional data space can be difficult to interpret, prompting the need for a more effective data representation. We address both of these issues by employing a complex network (graph) to represent climate data, a more intuitive model that can be used for analysis while also having a direct mapping to the physical world for interpretation. A cross correlation function is used to weight network edges, thus respecting the temporal nature of the data, and a community detection algorithm identifies multivariate clusters. Examining networks for consecutive periods allows us to study structural changes over time. We show that communities have a climatological interpretation and that disturbances in structure can be an indicator of climate events (or lack thereof). Finally, we discuss how this model can be applied for the discovery of more complex concepts such as unknown teleconnections or the development of multivariate climate indices and predictive insights.
Mining in a mobile environment.
nash.edu.au Wide-area sensor infrastructures, remote sensors, RFIDs, and wireless sensor networks yield massive volumes of disparate, dynamic, and geographically distributed data. As such sensors are becoming ubiquitous, a set of broad requirements is beginning to emerge across high-priority applications including adaptability to climate change, electric grid monitoring, disaster preparedness and management, national or homeland security, and the management of critical infrastructures. The raw data from sensors need to be efficiently managed and transformed to usable information through data fusion, which in turn must be converted to predictive insights via knowledge discovery, ultimately facilitating automated or human-induced tactical decisions or strategic policy based on decision sciences and decision support systems. Keeping in view the requirements of the emerging field of knowledge discovery from sensor data, we took initiative to develop a community of researchers with common interests and scientific goals, which culminated into the organization of SensorKDD series of workshops in conjunction with the
Adaptive Methods for Classification in Arbitrarily Imbalanced and Drifting Data Streams.

Model Monitor %28M2%29 Evaluating
This paper presents Model Monitor (M 2), a Java toolkit for robustly evaluating machine learning algorithms in the presence of changing data distributions. M 2 provides a simple and intuitive framework in which users can evaluate classifiers under hypothesized shifts in distribution and therefore determine the best model (or models) for their data under a number of potential scenarios. Additionally, M 2 is fully integrated with the WEKA machine learning environment, so that a variety of commodity classifiers can be used if desired.
A framework for monitoring classifiers%27 performance when and why failure occurs%3F
Abstract. Classifier error is the product of model bias and data variance. While understanding the bias involved when selecting a given learning algorithm, it is similarly important to understand the variability in data over time, since even the One True Model might perform poorly when training and evaluation samples diverge. Thus, the ability to identify distributional divergence is critical towards pinpointing when fracture points in classifier performance will occur, particularly since contemporary methods such as ten-folds and hold-out are poor predictors in divergent circumstances. This article implements a comprehensive evaluation framework to proactively detect breakpoints in classifiers ’ predictions and shifts in data distributions through a series of statistical tests. We outline and utilize three scenarios under which data changes: sample selection bias, covariate shift, and shifting class priors. We evaluate the framework with a variety of classifiers and datasets. 1.
Knowledge discovery from sensor data %28SensorKDD%29.
nash.edu.au Wide-area sensor infrastructures, remote sensors, RFIDs, and wireless sensor networks yield massive volumes of disparate, dynamic, and geographically distributed data. As such sensors are becoming ubiquitous, a set of broad requirements is beginning to emerge across high-priority applications including adaptability to climate change, electric grid monitoring, disaster preparedness and management, national or homeland security, and the management of critical infrastructures. The raw data from sensors need to be efficiently managed and transformed to usable information through data fusion, which in turn must be converted to predictive insights via knowledge discovery, ultimately facilitating automated or human-induced tactical decisions or strategic policy based on decision sciences and decision support systems. Keeping in view the requirements of the emerging field of knowledge discovery from sensor data, we took initiative to develop a community of researchers with common interests and scientific goals, which culminated into the organization of SensorKDD series of workshops in conjunction with the
Predicting individual disease risk based on medical history.
The monumental cost of health care, especially for chronic disease treatment, is quickly becoming unmanageable. This crisis has motivated the drive towards preventative medicine, where the primary concern is recognizing disease risk and taking action at the earliest signs. However, universal testing is neither time nor cost efficient. We propose CARE, a Collaborative Assessment and Recommendation Engine, which relies only on a patient’s medical history using ICD-9-CM codes in order to predict future diseases risks. CARE combines collaborative filtering methods with clustering to predict each patient’s greatest disease risks based on their own medical history and that of similar patients. We also describe an Iterative version, ICARE, which incorporates ensemble concepts for improved performance. These novel systems require no specialized information and provide predictions for medical conditions of all kinds in a single run. We present experimental results on a large Medicare dataset, demonstrating that CARE and ICARE perform well at capturing future disease risks. 1.
Learning Decision Trees for Unbalanced Data.
Abstract. Learning from unbalanced datasets presents a convoluted problem in which traditional learning algorithms may perform poorly. The objective functions used for learning the classifiers typically tend to favor the larger, less important classes in such problems. This paper compares the performance of several popular decision tree splitting criteria – information gain, Gini measure, and DKM – and identifies a new skew insensitive measure in Hellinger distance. We outline the strengths of Hellinger distance in class imbalance, propose its application in forming decision trees, and perform a comprehensive comparative analysis between each decision tree construction method. In addition, we consider the performance of each tree within a powerful sampling wrapper framework to capture the interaction of the splitting metric and sampling. We evaluate over this wide range of datasets and determine which operate best under class imbalance. 1
Troubleshooting thousands of jobs on production grids using data mining techniques.

Start Globally
Class imbalance is a ubiquitous problem in supervised learning and has gained wide-scale attention in the literature. Perhaps the most prevalent solution is to apply sampling to training data in order improve classifier performance. The typical approach will apply uniform levels of sampling globally. However, we believe that data is typically multi-modal, which suggests sampling should be treated locally rather than globally. It is the purpose of this paper to propose a framework which first identifies meaningful regions of data and then proceeds to find optimal sampling levels within each. This paper demonstrates that a global classifier trained on data locally sampled produces superior rank-orderings on a wide range of real-world and artificial datasets as compared to contemporary global sampling methods. 1
Data mining on the grid for the grid.
The Austrian Grid is the national Grid initiative in Austria to support grid computing in general, and to provide coordination and collaboration between research areas actively involved and interested in grid computing. The Austrian Grid Consortium combines Austria&#039;s leading researchers in advanced computing technologies with well-recognized partners in grid-dependent application areas. This is the third symposium organised by the Austrian Grid partners to demonstrate results of the cooperation (earlier symposia have been held in Hagenberg and Innsbruck). The symposium will be a forum for the institutions deploying the Austian grid infrastructure, for the Austrian developers of grid middleware extensions, and for the users of the Austrian Grid infrastructure to meet each other, share ideas and experience, and discuss future directions of the national grid efforts. Moreover, the symposium will gather an international research community that works on the development and application of state of the art Grid technology and reports on experience with Grid applications. Main topics of the symposium will be as follows: • Grid infrastructure deployment and management • Grid security and authentication services • Grid middleware services
Scaling up Classifiers to Cloud Computers.
As the size of available datasets has grown from Megabytes to Gigabytes and now into Terabytes, machine learning algorithms and computing infrastructures have continuously evolved in an effort to keep pace. But at large scales, mining for useful patterns still presents challenges in terms of data management as well as computation. These issues can be addressed by dividing both data and computation to build ensembles of classifiers in a distributed fashion, but trade-offs in cost, performance, and accuracy must be considered when designing or selecting an appropriate architecture. In this paper, we present an abstraction for scalable data mining that allows us to explore these tradeoffs. Data and computation are distributed to a computing cloud with minimal effort from the user, and multiple models for data management are available depending on the workload and system configuration. We demonstrate the performance and scalability characteristics of our ensembles using a wide variety of datasets and algorithms on a Condor-based pool with Chirp to handle the storage. 
Data mining on the grid for the grid.
The Austrian Grid is the national Grid initiative in Austria to support grid computing in general, and to provide coordination and collaboration between research areas actively involved and interested in grid computing. The Austrian Grid Consortium combines Austria&#039;s leading researchers in advanced computing technologies with well-recognized partners in grid-dependent application areas. This is the third symposium organised by the Austrian Grid partners to demonstrate results of the cooperation (earlier symposia have been held in Hagenberg and Innsbruck). The symposium will be a forum for the institutions deploying the Austian grid infrastructure, for the Austrian developers of grid middleware extensions, and for the users of the Austrian Grid infrastructure to meet each other, share ideas and experience, and discuss future directions of the national grid efforts. Moreover, the symposium will gather an international research community that works on the development and application of state of the art Grid technology and reports on experience with Grid applications. Main topics of the symposium will be as follows: • Grid infrastructure deployment and management • Grid security and authentication services • Grid middleware services
Analyzing PETs on Imbalanced Datasets When Training and Testing Class Distributions Differ.
Abstract. Many machine learning applications like finance, medicine, and risk management suffer from class imbalance: cases of interest occur rarely. Further complicating these applications is that the training and testing samples might differ significantly in their respective class distributions. Sampling has been shown to be a strong solution to imbalance and additionally offers a rich parameter space from which to select classifiers. This paper is concerned with the interaction between Probability Estimation Trees (PETs) [1], sampling, and performance metrics as testing distributions fluctuate substantially. A set of comprehensive analyses is presented, which anticipate classifier performance through a set of widely varying testing distributions. 1
Automatically countering imbalance and its empirical relationship to cost.
Automatically countering imbalance and its empirical
Knowledge discovery from sensor data %28SensorKDD%29.
nash.edu.au Wide-area sensor infrastructures, remote sensors, RFIDs, and wireless sensor networks yield massive volumes of disparate, dynamic, and geographically distributed data. As such sensors are becoming ubiquitous, a set of broad requirements is beginning to emerge across high-priority applications including adaptability to climate change, electric grid monitoring, disaster preparedness and management, national or homeland security, and the management of critical infrastructures. The raw data from sensors need to be efficiently managed and transformed to usable information through data fusion, which in turn must be converted to predictive insights via knowledge discovery, ultimately facilitating automated or human-induced tactical decisions or strategic policy based on decision sciences and decision support systems. Keeping in view the requirements of the emerging field of knowledge discovery from sensor data, we took initiative to develop a community of researchers with common interests and scientific goals, which culminated into the organization of SensorKDD series of workshops in conjunction with the
Using selective
We present a method to learn and recognize object class models from unlabeled and unsegmented cluttered scenes in a scale invariant manner. Objects are modeled as flexible constellations of parts. A probabilistic representation is used for all aspects of the object: shape, appearance, occlusion and relative scale. An entropy-based feature detector is used to select regions and their scale within the image. In learning the parameters of the scale-invariant object model are estimated. This is done using expectation-maximization in a maximum-likelihood setting. In recognition, this model is used in a Bayesian manner to classify images. The flexible nature of the model is demonstrated by excellent results over a range of datasets including geometrically constrained classes (e.g. faces, cars) and flexible objects (such as animals). 1.
Actively Exploring Creation of Face Space%28s%29 for Improved Face Recognition.

A Black%2DBox Approach to Query Cardinality Estimation.
We present a “black-box” approach to estimating query cardinality that has no knowledge of query execution plans and data distribution, yet provides accurate estimates. It does so by grouping queries into syntactic families and learning the cardinality distribution of that group directly from points in a high-dimensional input space constructed from the query’s attributes, operators, function arguments, aggregates, and constants. We envision an increasing need for such an approach in applications in which query cardinality is required for resource optimization and decision-making at locations that are remote from the data sources. Our primary case study is the Open SkyQuery federation of Astronomy archives, which uses a scheduling and caching mechanism at the mediator for execution of federated queries at remote sources. Experiments using real workloads show that the black-box approach produces accurate estimates and is frugal in its use of space and in computation resources. Also, the black-box approach provides dramatic improvements in the performance of caching in Open SkyQuery. 
Detecting Fractures in Classifier Performance.
A fundamental tenet assumed by many classification algorithms is the presumption that both training and testing samples are drawn from the same distribution of data – this is the stationary distribution assumption. This entails that the past is strongly indicative of the future. However, in real world applications, many factors may alter the One True Model responsible for generating the data distribution both significantly and subtly. In circumstances violating the stationary distribution assumption, traditional validation schemes such as ten-folds and hold-out become poor performance predictors and classifier rankers. Thus, it becomes critical to discover the fracture points in classifier performance by discovering the divergence between populations. In this paper, we implement a comprehensive evaluation framework to identify bias, enabling selection of a ”correct ” classifier given the sample bias. To thoroughly evaluate the performance of classifiers within biased distributions, we consider the following three scenarios: missing completely at random (akin to stationary); missing at random; and missing not at random. The latter reflects the canonical sample selection bias problem. 1
Enhanced Situational Awareness Application of DDDAS Concepts to Emergency and Disaster Management.
Abstract. We describe a prototype emergency and disaster information system designed and implemented using DDDAS concepts. The system is designed to use real-time cell phone calling data from a geographical region, including calling activity – who calls whom, call duration, services in use, and cell phone location information – to provide enhanced situational awareness for managers in emergency operations centers (EOCs) during disaster events. Powered-on cell phones maintain contact with one or more within-range cell towers so as to receive incoming calls. Thus, location data about all phones in an area are available, either directly from GPS equipped phones, or by cell tower, cell sector, distance from tower and triangulation methods. This permits the cell phones of a geographical region to serve as an ad hoc mobile sensor net, measuring the movement and calling patterns of the population. A prototype system, WIPER, serves as a test bed to research open DDDAS design issues, including dynamic validation of simulations, algorithms to interpret high volume data streams, ensembles of simulations, runtime execution, middleware services, and experimentation frameworks [1]. 1
Exploiting Diversity in Ensembles Improving the Performance on Unbalanced Datasets.
Abstract. Ensembles are often capable of greater predictive performance than any of their individual classifiers. Despite the need for classifiers to make different kinds of errors, the majority voting scheme, typically used, treats each classifier as though it contributed equally to the group‘s performance. This can be particularly limiting on unbalanced datasets, as one is more interested in complementing classifiers that can assist in improving the true positive rate without signicantly increasing the false positive rate. Therefore, we implement a genetic algorithm based framework to weight the contribution of each classifier by an appropriate fitness function, such that the classifiers that complement each other on the unbalanced dataset are preferred, resulting in significantly improved performances. The proposed framework can be built on top of any collection of classifiers with different fitness functions. 1
Authentication anomaly detection a case study on a virtual private network.

Anomaly detection in a mobile communication network.
Cell phone networks produce a massive volume of service usage data which, when combined with location data, can be used to pinpoint emergency situations that cause changes in network usage. Such a change may be the results of an increased number of people trying to call friends or family to tell them what is happening or a decrease in network usage caused by people being unable to use the network. Such events are anomalies and managing emergencies effectively requires identifying anomalies quickly. This problem is difficult due to the rate at which very large volumes of data are produced. In this paper, we discuss the use of data stream clustering algorithms for anomaly detection. Contact:
Combating imbalance in network intrusion datasets.

Troubleshooting Distributed Systems via Data Mining.

Evolutionary Ensemble Creation and Thinning.

Evaluation of Summarization Schemes for Learning in Streams.

Data management and query %2D Estimating query result sizes for proxy caching in scientific database federations.

Data management and query %2D Estimating query result sizes for proxy caching in scientific database federations.

A study in machine learning from imbalanced data for sentence boundary detection in speech.
Enriching speech recognition output with sentence boundaries improves its human readability and enables further processing by downstream language processing modules. We have constructed a hidden Markov model (HMM) system to detect sentence boundaries that uses both prosodic and textual information. Since there are more nonsentence boundaries than sentence boundaries in the data, the prosody model, which is implemented as a decision tree classifier, must be constructed to effectively learn from the imbalanced data distribution. To address this problem, we investigate a variety of sampling approaches and a bagging scheme. A pilot study was carried out to select methods to apply to the full NIST sentence boundary evaluation task across two corpora (conversational telephone speech and broadcast news speech), using both human transcriptions and recognition output. In the pilot study, when classification error rate is the performance measure, using the original training set achieves the best performance among the sampling methods, and an ensemble of multiple classifiers from different downsampled training sets achieves
Random Subspaces and Subsampling for 2%2DD Face Recognition.

Many Are Better Than One Improving Probabilistic Estimates from Decision Trees.

Designing Multiple Classifier Systems for Face Recognition.

Data Mining for Imbalanced Datasets An Overview.

Learning From Labeled And Unlabeled Data An Empirical Study Across Techniques And Domains.
©2005 AI Access Foundation. All rights reserved. There has been increased interest in devising learning techniques that combine unlabeled data with labeled data – i.e. semi-supervised learning. However, to the best of our knowledge, no study has been performed across various techniques and different types and amounts of labeled and unlabeled data. Moreover, most of the published work on semi-supervised learning techniques assumes that the labeled and unlabeled data come from the same distribution. It is possible for the labeling process to be associated with a selection bias such that the distributions of data points in the labeled and unlabeled sets are different. Not correcting for such bias can result in biased function approximation with potentially poor performance. In this paper, we present an empirical study of various semi-supervised learning techniques on a variety of datasets. We attempt to answer various questions such as the effect of independence or relevance amongst features, the effect of the size of the labeled and unlabeled sets and the effect of noise. We also investigate the impact of sample-selection bias on the semi-supervised learning techniques under study and implement a bivariate probit technique particularly designed to correct for such bias. 1.
Classification and knowledge discovery in protein databases.

Learning Ensembles from Bites A Scalable and Accurate Approach.
Bagging and boosting are two popular ensemble methods that typically achieve better accuracy than a single classifier. These techniques have limitations on massive datasets, as the size of the dataset can be a bottleneck. Voting many classifiers built on small subsets of data (&#034;pasting small votes&#034;) is a promising approach for learning from massive datasets, one that can utilize the power of boosting and bagging. We propose a framework for building hundreds or thousands of such classifiers on small subsets of data in a distributed environment. Experiments show this approach is fast, accurate, and scalable.
Editorial special issue on learning from imbalanced data sets.

SMOTEBoost Improving Prediction of the Minority Class in Boosting.
Abstract. Many real world data mining applications involve learning from imbalanced data sets, where the particular events of interest may be very few when compared to the other classes. Learning from data sets that contain rare events usually produces biased classifiers that have a higher predictive accuracy over the majority class(es), but poorer predictive accuracy over the minority class of interest. SMOTE (Synthetic Minority Over-sampling Technique) is a recent approach that is specifically designed for learning with minority classes. Boosting is a promising ensemble-based learning algorithm that can improve the classification performance of any weak classifier. This paper presents a novel approach for learning from rare classes, based on a combination of the SMOTE algorithm and the boosting procedure. Unlike standard boosting where all misclassified examples are given equal weights, the novel SMOTEBoost approach creates synthetic examples from the minority class, thus indirectly changing the updating weights and compensating for skewed distributions. The SMOTEBoost algorithm applied to several highly and moderately imbalanced data sets shows improvement in prediction performance on the rare class compared to standard boosting and to the SMOTE algorithm used with a single classifier. 1.
Distributed learning with bagging%2Dlike performance.
Bagging forms a committee of classifiers by bootstrap aggregation of training sets from a pool of training data. A simple alternative to bagging is to partition the data into disjoint subsets. Experiments with decision tree and neural network classifiers on various datasets show that, given the same size partitions and bags, disjoint partitions result in performance equivalent to, or better than, bootstrap aggregates (bags). Many applications (e.g., protein structure prediction) involve use of datasets that are too large to handle in the memory of the typical computer. Hence, bagging with samples the size of the data is impractical. Our results indicate that, in such applications, the simple approach of creating a committee of n classifiers from disjoint partitions each of size 1/n (which will be memory resident during learning) in a distributed way results in a classifier which has a bagging-like performance gain. The use of distributed disjoint partitions in learning is significantly less complex and faster than bagging.
Generalization Methods in Bioinformatics.
Protein secondary structure prediction and high-throughput drug screen data mining are two important applications in bioinformatics. The data is represented in sparse feature spaces and can be unrepresentative of future data. Supervised learners in this context will display their inherent bias toward certain solutions, generally solutions that fit the training set well. In this paper, we first describe an ensemble approach using subsampling that scales well with dataset size. A sufficient number of ensemble members using subsamples of the data can yield a more accurate classifier than a single classifier using the entire dataset. Experiments on several datasets demonstrate the effectiveness of the approach. We report results from the KDD Cup 2001 drug discovery dataset in which our approach yields a higher weighted accuracy than the winning entry. We then extend our ensemble approach to create an over-generalized classifier for prediction by reducing the individual subsample size. The ensemble strategy using small subsamples has the effect of averaging over a wider range of hypotheses. Weshow that both protein secondary structure prediction and drug discovery prediction can be improved by the use of overgeneralization, specifically through the use of ensembles of small subsamples.
Distributed Pasting of Small Votes.
Bagging and boosting are two popular ensemble methods that achieve better accuracy than a single classifier. These techniques have limitations on massive datasets, as the size of the dataset can be a bottleneck. Voting many classifiers built on small subsets of data (&#034;pasting small votes&#034;) is a promising approach for learning from massive datasets. Pasting small votes can utilize the power of boosting and bagging, and potentially scale up to massive datasets. We propose a framework for building hundreds or thousands of such classifiers on small subsets of data in a distributed environment. Experiments show this approach is fast, accurate, and scalable to massive datasets.
SMOTE Synthetic Minority Over%2Dsampling Technique.
An approach to the construction of classifiers from imbalanced datasets is described. A dataset is imbalanced if the classification categories are not approximately equally represented. Often real-world data sets are predominately composed of ``normal&#039;&#039; examples with only a small percentage of ``abnormal&#039;&#039; or ``interesting&#039;&#039; examples. It is also the case that the cost of misclassifying an abnormal (interesting) example as a normal example is often much higher than the cost of the reverse error. Under-sampling of the majority (normal) class has been proposed as a good means of increasing the sensitivity of a classifier to the minority class. This paper shows that a combination of our method of over-sampling the minority (abnormal) class and under-sampling the majority (normal) class can achieve better classifier performance (in ROC space) than only under-sampling the majority class. This paper also shows that a combination of our method of over-sampling the minority class and under-sampling the majority class can achieve better classifier performance (in ROC space) than varying the loss ratios in Ripper or class priors in Naive Bayes. Our method of over-sampling the minority class involves creating synthetic minority class examples. Experiments are performed using C4.5, Ripper and a Naive Bayes classifier. The method is evaluated using the area under the Receiver Operating Characteristic curve (AUC) and the ROC convex hull strategy.
Investigation of bagging%2Dlike effects and decision trees versus neural nets in protein secondary structure prediction.

Bagging Is a Small%2DData%2DSet Phenomenon.
Bagging forms a committee of classifiers by bootstrap aggregation of training sets from a pool of training data. A simple alternative to bagging is to partition the data into disjoint subsets. Experiments on various datasets show that, given the same size partitions and bags, the use of disjoint partitions results in better performance than the use of bags. Many applications (e.g., protein structure prediction) involve the use of datasets that are too large to handle in the memory of the typical computer. Our results indicate that, in such applications, the simple approach of creating a committee of classifiers from disjoint partitions is to be preferred over the more complex approach of bagging.
Creating Ensembles of Classifiers.
Ensembles of classifiers offer promise in increasing overall classification accuracy. Combining multiple
Learning Rules from Distributed Data.
In this paper a concern about the accuracy (as a function of parallelism) of a certain class of distributed learning algorithms is raised, and one proposed improvement is illustrated. We focus on learning a single model from a set of disjoint data sets, which are distributed across a set of computers. The model is a set of rules. The distributed data sets may be disjoint for any of several reasons. In our approach, the first step is to construct a rule set (model) for each of the original disjoint data sets. Then rule sets are merged until an eventual final rule set is obtained which models the aggregate data. We show that this approach compares to directly creating a rule set from the aggregate data and promises faster learning. Accuracy can drop off as the degree of parallelism increases. However, an approach has been developed to reduce this problem.
