Oversampling Method Research Articles

A Novel Over-Sampling Method Based on EDAs for Learning from Imbalanced Data Sets

A Novel Over-Sampling Method Based on EDAs for Learning from Imbalanced Data Sets

A Comparison Study of Cost-Sensitive Learning and Sampling Methods on Imbalanced Data Sets

The classifier, built from a highly-skewed class distribution data set, generally predicts an unknown sample as the majority class much more frequently than the minority class. This is due to the fact that the aim of classifier is designed to get the highest classification accuracy. We compare three classification methods dealing with the data sets in which class distribution is imbalanced and has non-uniform misclassification cost, namely cost-sensitive learning method whose misclassification cost is embedded in the algorithm, over-sampling method and under-sampling method. In this paper, we compare these three methods to determine which one will produce the best overall classification under any circumstance. We have the following conclusion: 1. Cost-sensitive learning is suitable for the classification of imbalanced dataset. It outperforms sampling methods overall, and is more stable than sampling methods except the condition that data set is quite small. 2. If the dataset is highly skewed or quite small, over-sampling methods may be better.

Optimal sampling from sliding windows

A sliding windows model is an important case of the streaming model, where only the most “recent” elements remain active and the rest are discarded. The sliding windows model is important for many applications (see, e.g., Babcock, Babu, Datar, Motwani and Widom (PODS 02); and Datar, Gionis, Indyk and Motwani (SODA 02)). There are two equally important types of the sliding windows model – windows with fixed size (e.g., where items arrive one at a time, and only the most recent n items remain active for some fixed parameter n), and timestamp-based windows (e.g., where many items can arrive in “bursts” at a single step and where only items from the last t steps remain active, again for some fixed parameter t). Random sampling is a fundamental tool for data streams, as numerous algorithms operate on the sampled data instead of on the entire stream. Effective sampling from sliding windows is a nontrivial problem, as elements eventually expire. In fact, the deletions are implicit; i.e., it is not possible to identify deleted elements without storing the entire window. The implicit nature of deletions on sliding windows does not allow the existing methods (even those that support explicit deletions, e.g., Cormode, Muthukrishnan and Rozenbaum (VLDB 05); Frahling, Indyk and Sohler (SOCG 05)) to be directly “translated” to the sliding windows model. One trivial approach to overcoming the problem of implicit deletions is that of over-sampling. When k samples are required, the over-sampling method maintains k ′ > k samples in the hope that at least k samples are not expired. The obvious disadvantages of this method are twofold: (a) It introduces additional costs and thus decreases the performance; and (b) The memory bounds are not deterministic, which is atypical for streaming algorithms (where even small probability events may eventually happen for a stream that is long enough). Babcock, Datar and Motwani (SODA 02), were the first to stress the importance of improvements to over-sampling. They formally introduced the problem of sampling from sliding windows and improved the over-sampling method for sampling with replacement. Their elegant solutions for sampling with replacement are optimal in expectation, and thus resolve disadvantage (a) mentioned above. Unfortunately, the randomized bounds do not resolve disadvantage (b) above. Interestingly, all algorithms that employ the ideas of Babcock, Datar and Motwani have the same central problem of having to deal with a randomized complexity bound (see, e.g., Datar and Muthukrishnan (ESA 02); Chakrabarti, Cormode and McGregor (SODA 07)). Further, the proposed solutions of Babcock, Datar and Motwani for sampling without replacement are based on the over-sampling method and thus do not solve problem (a). Therefore, the question of whether we can solve sampling on sliding windows optimally (i.e., resolve both disadvantages) is implicit in the paper of Babcock, Datar and Motwani and has remained open for all variants of the problem. In this paper we answer these questions affirmatively and provide optimal sampling schemas for all variants of the problem, i.e., sampling with or without replacement from fixed or timestamp-based windows. Specifically, for fixed-size windows, we provide optimal solutions that require O ( k ) memory; for timestamp-based windows, we show algorithms that require O ( k log n ) space, 1 which is optimal since it matches the lower bound by Gemulla and Lehner (SIGMOD 08). In contrast to the work of Babcock, Datar and Motwani, our solutions have deterministic bounds.

Borderline over-sampling for imbalanced data classification

Traditional classification algorithms usually provide poor accuracy on the prediction of the minority class of imbalanced data sets. This paper proposes a new method for dealing with imbalanced data sets by over-sampling the borderline minority class instances. A Support Vector Machine (SVM) classifier is then trained to predict future instances. Compared with other over-sampling methods, the proposed method focuses only on the minority class instances residing along the decision boundary, due to the fact that this region is the most crucial for establishing the decision boundary. Furthermore, the artificial minority instances are generated in such a way that the regions of the minority class with fewer majority class instances would be expanded by extrapolation, otherwise the current boundary of the minority class would be consolidated by interpolation. Experimental results show that the proposed method achieves a better performance than other over-sampling methods.

Lots of Aloha, Little Data: Data and Research on Native Hawaiian and Pacific Islanders

Knowledge is power in social change efforts, and data and diverse approaches are needed to fuel it. Recommendations from this section's experts in the field are united in calling for using Office of Management and Budget (OMB) guidelines consistently; disaggregating data collection and reporting on socioeconomic, educational, cultural, youth, and health challenges facing Native Hawaiian and Pacific Islanders (NHPIs); improving data estimates by addressing small sample sizes with oversampling methods and targeted surveys; and collaborating with public and private community-based organizations to promote high response rates and/or data partnerships. A great need exists for trend analyses over time on NHPIs. These analyses will shed insight on the effects of programs and policies, including longitudinal studies, such as the ECLS and the National Longitudinal Survey of Youth. Consistently, research by these authors highlights the power of culturally relevant approaches and the importance of developing innovative approaches to data, knowledge, and research that embrace indigenous perspectives in program and policy design, particularly in education, health, and science.

Lots of Aloha, Little Data: Data and Research on Native Hawaiian and Pacific Islanders

Knowledge is power in social change efforts, and data and diverse approaches are needed to fuel it. Recommendations from this section's experts in the field are united in calling for using Office of Management and Budget (OMB) guidelines consistently; disaggregating data collection and reporting on socioeconomic, educational, cultural, youth, and health challenges facing Native Hawaiian and Pacific Islanders (NHPIs); improving data estimates by addressing small sample sizes with oversampling methods and targeted surveys; and collaborating with public and private community-based organizations to promote high response rates and/or data partnerships. A great need exists for trend analyses over time on NHPIs. These analyses will shed insight on the effects of programs and policies, including longitudinal studies, such as the ECLS and the National Longitudinal Survey of Youth. Consistently, research by these authors highlights the power of culturally relevant approaches and the importance of developing innovat...

An Iterative Peak Power Reduction Method with Adaptive Intermediary Over-Sampling for Wireless OFDMA Systems

This paper proposes an iterative peak power reduction method with adaptive intermediary over-sampling which uses the necessary minimum bandwidth according to iteration number for wireless OFDMA systems. The required bandwidth to each iteration number is evaluated by computer simulation, and over-sampling numbers in iterative processing are controlled by using the simulation results. The results show that the required bandwidth is 1.6, 2.0, and 2.7 times of the used signal bandwidth at the iteration number of 1, 2, and 3, respectively. The proposed adaptive over-sampling method can reduce its multiplication number by 13%.

Mix-ratio sampling: Classifying multiclass imbalanced mouse brain images using support vector machine

Support Vector Machine (SVM) is a classifier designed to achieve optimized classification accuracy. It has been applied to numerous applications associated with images. Yet challenges remain when applying SVM on segmenting mouse brain images. This is due to the fact that each high-resolution mouse brain image is a very large data set and it is a multiclass classification problem with extremely imbalanced data size for different classes. To address these issues, a mix-ratio sampling approach for SVM is proposed which determines various over-sampling ratios for different minority classes. In addition, to improve the imaging classification accuracy, spatial information is incorporated into the classification problem. Five mouse Magnetic Resonance Microscopy (MRM) images are collected to test the accuracy of classifying 21 brain structures. The first comparison experiment demonstrates the SVM with mix-ratio sampling method relieves the imbalance problem for multiclass more effectively and efficiently than the SVM with simple over-sampling method. In the second comparison experiment, another classifier, Artificial Neural Network (ANN) is used to compare against SVM based on the same mix-ratio sampled data and the results indicate that SVM shows better classification performance than ANN. Thirdly, the cross validation is conducted to demonstrate SVM with mix-ration sampling can classify multiclass imbalanced data with high accuracy.

A Data-Driven Approach to Manage the Length of Stay for Appendectomy Patients

Skyrocketing patient-care costs demand that health-care institutions improve their resource-utilization effectiveness and efficiency. The length of an inpatient's stay has direct significant impacts on patient-care costs, service quality, and outcomes. Despite attempts to manage the length of stay (LOS) for frequently performed surgical procedures (e.g., appendectomies), many service providers cannot achieve the target range allowed by the managed care system. We take a data-driven approach to predict which appendectomy patients will likely have a LOS beyond that reimbursable by the underlying managed care system. We use a support vector machine to construct a generic prediction system and then extend that system by incorporating a resampling or cost-sensitive method to address the imbalanced sample problem. Using 557 appendectomy cases from a tertiary medical center in Taiwan, we examine the effectiveness of the generic prediction system compared with the effectiveness of its extensions. The results suggest the viability of a data-driven approach to manage LOS by enabling service providers to identify in advance those patients who will likely need extended stays. The comparative analyses also show the relative advantages of the oversampling and cost-sensitive methods for addressing the imbalanced sample problem. The findings have important implications for research and practice.

A comparative study for spatial resolution and subjective image characteristics of a multi-slice CT and a cone-beam CT for dental use

Purpose Multi-slice CT (MSCT) and cone-beam CT (CBCT) are widely used in dental practice. This study compared the spatial resolution of these CT systems to elucidate which CT modalities should be selected for various clinical cases. Materials and methods As MSCT and CBCT apparatuses, Somatom Sensation 64 and 3D Accuitomo instruments, respectively, were used. As an objective evaluation of spatial resolution of these CT systems, modulation transfer function (MTF) analysis was performed employing an over-sampling method. The results of MTF analysis were confirmed with a line-pair test using CATPHAN. As a subjective evaluation, a microstructure visualization ability study was performed using a Jcl:SD rat and a head CT phantom. Results MTF analysis showed that for the in-plane direction, the z-axis ultrahigh resolution mode (zUHR) of the Sensation 64 and 3D Accuitomo instruments had higher spatial resolutions than the conventional mode (64×) of the Sensation 64, but for the longitudinal direction, the 3D Accuitomo had clearly higher spatial resolution than either mode of the Sensation 64. A line-pair test study and microstructure visualization ability studies confirmed the results for MTF analysis. However, images of the rat and the CT phantom revealed that the 3D Accuitomo demonstrated the failure to visualize the soft tissues along with aliasing and beam-hardening artifacts, which were not observed in the Sensation 64. Conclusions This study successfully applied spatial resolution analysis using MSCT and CBCT systems in a comparative manner. These findings could help in deciding which CT modality should be selected for various clinical cases.

Flow based oversampling technique for multiscale finite element methods

Oversampling techniques are often used in porous media simulations to achieve high accuracy in multiscale simulations. These methods reduce the effect of artificial boundary conditions that are imposed in computing local quantities, such as upscaled permeabilities or basis functions. In the problems without scale separation and strong non-local effects, the oversampling region is taken to be the entire domain. The basis functions are computed using single-phase flow solutions which are further used in dynamic two-phase simulations. The standard oversampling approaches employ generic global boundary conditions which are not associated with actual flow boundary conditions. In this paper, we propose a flow based oversampling method where the actual two-phase flow boundary conditions are used in constructing oversampling auxiliary functions. Our numerical results show that the flow based oversampling approach is several times more accurate than the standard oversampling method. We provide partial theoretical explanation for these numerical observations.

Machine learning approaches to supporting the identification of photoreceptor-enriched genes based on expression data

BackgroundRetinal photoreceptors are highly specialised cells, which detect light and are central to mammalian vision. Many retinal diseases occur as a result of inherited dysfunction of the rod and cone photoreceptor cells. Development and maintenance of photoreceptors requires appropriate regulation of the many genes specifically or highly expressed in these cells. Over the last decades, different experimental approaches have been developed to identify photoreceptor enriched genes. Recent progress in RNA analysis technology has generated large amounts of gene expression data relevant to retinal development. This paper assesses a machine learning methodology for supporting the identification of photoreceptor enriched genes based on expression data.ResultsBased on the analysis of publicly-available gene expression data from the developing mouse retina generated by serial analysis of gene expression (SAGE), this paper presents a predictive methodology comprising several in silico models for detecting key complex features and relationships encoded in the data, which may be useful to distinguish genes in terms of their functional roles. In order to understand temporal patterns of photoreceptor gene expression during retinal development, a two-way cluster analysis was firstly performed. By clustering SAGE libraries, a hierarchical tree reflecting relationships between developmental stages was obtained. By clustering SAGE tags, a more comprehensive expression profile for photoreceptor cells was revealed. To demonstrate the usefulness of machine learning-based models in predicting functional associations from the SAGE data, three supervised classification models were compared. The results indicated that a relatively simple instance-based model (KStar model) performed significantly better than relatively more complex algorithms, e.g. neural networks. To deal with the problem of functional class imbalance occurring in the dataset, two data re-sampling techniques were studied. A random over-sampling method supported the implementation of the most powerful prediction models. The KStar model was also able to achieve higher predictive sensitivities and specificities using random over-sampling techniques.ConclusionThe approaches assessed in this paper represent an efficient and relatively inexpensive in silico methodology for supporting large-scale analysis of photoreceptor gene expression by SAGE. They may be applied as complementary methodologies to support functional predictions before implementing more comprehensive, experimental prediction and validation methods. They may also be combined with other large-scale, data-driven methods to facilitate the inference of transcriptional regulatory networks in the developing retina. Furthermore, the methodology assessed may be applied to other data domains.

Three-Dimensional Aberration-Free Imaging Recovery by the Oversampling Method and Tomography

Extract HTML view is not available for this content. However, as you have access to this content, a full PDF is available via the ‘Save PDF’ action button. Extended abstract of a paper presented at Microscopy and Microanalysis 2005 in Honolulu, Hawaii, USA, July 31--August 4, 2005

A study of the behavior of several methods for balancing machine learning training data

There are several aspects that might influence the performance achieved by existing learning systems. It has been reported that one of these aspects is related to class imbalance in which examples in training data belonging to one class heavily outnumber the examples in the other class. In this situation, which is found in real world data describing an infrequent but important event, the learning system may have difficulties to learn the concept related to the minority class. In this work we perform a broad experimental evaluation involving ten methods, three of them proposed by the authors, to deal with the class imbalance problem in thirteen UCI data sets. Our experiments provide evidence that class imbalance does not systematically hinder the performance of learning systems. In fact, the problem seems to be related to learning with too few minority class examples in the presence of other complicating factors, such as class overlapping. Two of our proposed methods deal with these conditions directly, allying a known over-sampling method with data cleaning methods in order to produce better-defined class clusters. Our comparative experiments show that, in general, over-sampling methods provide more accurate results than under-sampling methods considering the area under the ROC curve (AUC). This result seems to contradict results previously published in the literature. Two of our proposed methods, Smote + Tomek and Smote + ENN, presented very good results for data sets with a small number of positive examples. Moreover, Random over-sampling, a very simple over-sampling method, is very competitive to more complex over-sampling methods. Since the over-sampling methods provided very good performance results, we also measured the syntactic complexity of the decision trees induced from over-sampled data. Our results show that these trees are usually more complex then the ones induced from original data. Random over-sampling usually produced the smallest increase in the mean number of induced rules and Smote + ENN the smallest increase in the mean number of conditions per rule, when compared among the investigated over-sampling methods.

An advanced sector servo using a learning, estimating and over-sampling method for a 3.5" 28 Mbyte FDD

The authors developed an advanced sector servo which was used in a 28-Mbyte flexible magnetic disk drive (FDD). This sector servo uses a learning method which estimates a continuous track position error between servo sectors based on the information from an external position encoder and an over-sampling method which samples position error at a higher sampling rate. As a result, the authors achieved fine-trackability and high vibration-resistance performance without an increase in the number of servo sectors. The head positioning control system using these methods and the control performances are described. >

Hardware-assisted polygon antialiasing

The use of two sampling methods, the precise point-sampling algorithm and the extended Bresenham algorithm (EBA), for rendering hardware is discussed, showing where the two methods converge and diverge. Two antialiasing techniques are then presented: an oversampling method and a stochastic method. Practical application of both is described. The use of filtering and sophisticated features with these methods is considered. Hardware support for antialiasing of polygons is discussed. >

ズームＦＦＴを用いたオーバサンプリング法とかみあい誤差分析への応用

This paper offers an over-sampling method combined with zoom-FFT. The method is one of the methods which increases the resolving power of frequency in FFT analysis, and has the feature that it can choose arbitrary zoom ratio, i.e., arbitrary magnification of the resolving power of frequency. The results of investigation on the method are as follows : With the zoom ratio which make the fraction of normalized frequency some integer, the frequency component can be exactly extracted on its correct position; Smallest zoom ratio is sufficient for analysis; For Hamming window, the resolving power of two components is 2-normalized frequency under a zoom ratio smaller than 30; If the distance between two components is larger than 6-normalized frequency, Hanning window gives exact analysis with an amplitude error smaller than 0.1% and a phase error smaller than 0.1 degrees.