Outliers Research Articles

Metric-Based Semi-Supervised Regression

Regression problems are present in many industrial applications, and many supervised learning algorithms have been devised over decades. However, available labeled examples are limited in some application settings; meanwhile, enormous unlabeled examples are relatively easy to collect. Thus, this work proposes a simple but effective method to cope with semi-supervised regression problems. We propose to use deep neural networks to develop our proposed method as deep learning has shown promising results in recent years. Our proposed method is a metric-based approach, and the goal is to learn an embedding space by metric learning with few labeled examples and enormous unlabeled examples. The regression estimation of the target data point is performed on the new space. We generate an artificial dataset based on several criteria to investigate whether the proposed model could make accurate predictions on the data samples that have specific properties. The experimental results point that our proposed model could capture the trend of a non-linear function and normally predict well even though this dataset comprises extreme outliers. Moreover, we conduct experiments on four datasets and compare our proposed work with several alternatives. The experimental results indicate that our proposed method achieves promising results. Besides performance evaluation, detailed analysis about our proposed method is also provided in this work.

Radiation dose in cone beam CT guided ERCP

PurposeTo evaluate the practical and radiation safety considerations in utilization of 3-D imaging through cone beam computed tomography (CBCT) in endoscopic retrograde cholangiopancreatography (ERCP). MethodERCP procedures were performed using an Artis Q interventional suite (Siemens Healthineers), including 42 procedures with CBCT. Some CBCT cases used the standard “DR” exposure protocol and some used the low dose protocol “DR care”. Data on x-ray radiation doses were retrospectively collected and to help optimize technical factors surrounding the procedure and compared to radiation exposure data of patients undergoing conventional ERCP. ResultsThe median dose area product was 24.4 Gycm2 for one DR volume and 5.07 Gycm2 for one DR care volume. The median total dose area product was 6.52 Gycm2 for conventional ERCP, 48.9 Gycm2 for procedures using DR and 19.7 Gycm2 when using DR care. Conventional ERCP resulted in a significantly lower radiation dose than procedures using either CBCT protocols (p < 0.001). However, conventional ERCP showed a large number of outliers with higher dose at the level of, or surpassing, CBCT procedures. ConclusionsFor less complex procedures with small x-ray doses and short fluoroscopy times conventional 2-D x-ray technique is clinically sufficient and utilizes less radiation dose than CBCT methods, but there is a wide range of doses and extreme outliers have been observed. CBCT can facilitate ERCP in cases with difficult ducal anatomy, possibly reducing procedural time, x-ray exposure time and need for contrast injections. Methods for prediction on high radiation ERCPs are needed to standardize and optimize patient selection for CBCT-ERCP.

Prediction of anaerobic digestion performance and identification of critical operational parameters using machine learning algorithms

Machine learning has emerges as a novel method for model development and has potential to be used to predict and control the performance of anaerobic digesters. In this study, several machine learning algorithms were applied in regression and classification models on digestion performance to identify determinant operational parameters and predict methane production. In the regression models, k-nearest neighbors (KNN) algorithm demonstrates optimal prediction accuracy (root mean square error = 26.6, with the dataset range of 259.0–573.8), after narrowing prediction coverage by excluding extreme outliers from the validation set. In the classification models, logistic regression multiclass algorithm yields the best prediction accuracy of 0.73. Feature importance reveals that total carbon was the determinant operational parameter. These results demonstrate the great potential of using machine learning algorithms to predict anaerobic digestion performance.

Prediction of compression strength of ancient mortars through in situ drilling resistance technique

The paper presents the results of a research project carried out to determine the mechanical characteristics of mortar in historical masonry buildings by drilling resistance technique (DRMS).A set of mortar mixtures have been produced in experimental campaigns covering different compressive strengths typical of historic mortars. The mortars have been obtained with different classes of binder and river sand with different grading curves. Specimens obtained were subjected both to flexural, compression and DRMS laboratory tests.Considering the results of the drilling measurement distributions, both the mode and the average value, the latter obtained after a statistical treatment in which extreme outliers are deleted, have been taken as reference values for each test performed. Effective correlations between the mode and average values of DRMS tests and the compressive strengths of the mortars are provided.Finally, a stepwise process to be followed for in situ use of DRMS is defined and applied to mortar joints of some existing masonry buildings in Tuscany (Italy).

Robust asset allocation with conditional value at risk using the forward search

AbstractThe well‐known Markowitz approach to portfolio allocation, based on expected returns and their covariance, seems to provide questionable results in financial management. One motivation for the pitfall is that financial returns have heavier than Gaussian tails, so the covariance of returns, used in the Markowitz model as a measure of portfolio risk, is likely to provide a loose quantification of the effective risk. Additionally, the Markowitz approach is very sensitive to small changes in either the expected returns or their correlation, often leading to irrelevant portfolio allocations. More recent allocation techniques are based on alternative risk measures, such as value at risk (VaR) and conditional VaR (CVaR), which are believed to be more accurate measures of risk for fat‐tailed distributions. Nevertheless, both VaR and CVaR estimates can be influenced by the presence of extreme returns. In this paper, we discuss sensitivity to the presence of extreme returns and outliers when optimizing the allocation, under the constraint of keeping CVaR to a minimum. A robust and efficient approach, based on the forward search, is suggested. A Monte Carlo simulation study shows the advantages of the proposed approach, which outperforms both robust and nonrobust alternatives under a variety of specifications. The performance of the method is also thoroughly evaluated with an application to a set of US stocks.

Application of Geostatistics for Cleat Data Validation in Sungai Lilin, Musi Banyuasin Region, South Sumatra

Cleat on coal seam is a slope stability controller, as well as permeability and porosity for fluidal flows. It also has become an interesting issue on coal mining, nevertheless the finite quantitative data. As a matter of facts, there is actually some available data about cleat as attributes, but this study only objecting on the orientation, cleat spacing, and cleat aperture. Cleat attributes measurement through field observation at two coalfields;Baturona and PMC in the area of Sungai Lilin, MusiBanyuasin Region. Next, those 667 faces and butt cleats were measured from 8 scanlines being analyzed using geostatistic methods. The purpose of geostatisticanalysis in this study is to see the relation between variables (cleat attributes) to evaluate data quantitatively until the normal distribution is obtained. The data filter was done as a treatment until valid data is reached through the boxplot model. Early boxplot result that almost entirely had outliers data in each seam. In fact, it is commonly seen some extreme outlier data. This suggests that outlier data filter is treated as a treatment until a normal distribution system is achieved on each cleat attribute. Identification of cleat outliers has been affected by mining activity, or due to weathering processes in the coal seams that have been exposed.

Validation of a Motor-Cognitive Assessment for a Stepping Exergame in Older Adults: Use of Game-Specific, Internal Data Stream.

Objective: To comprehensively validate an assessment tailored to an exergaming training program with motor-cognitive challenges. Materials and Methods: Fifty-eight cognitively intact, multimorbid, independently living older adults (mean age [standard deviation]: 78.3 [6.5] years) with moderate functional limitations participated in the study. For construct validity, Spearman's rank correlations (rs) between game parameters and established cognitive (Number-Connection-Test [Zahlen-Verbindungs-Test], Simple Response Time Task, Simon Task) and motor (Short Physical Performance Battery, Physiomat® balance tests) measures were calculated. Test-retest reliability was documented by intraclass correlation coefficients (ICCs), sensitivity to change by effect sizes using partial eta squared (ηp2), and feasibility by mean completion time and completion rates. Results: Good construct validity of the assessment was observed, with on average moderate-to-high correlations between game parameters and cognitive tests, measures of lower extremity function and dynamic balance (range of rs including extreme outliers = 0.00-0.70, P < 0.001-0.998). Test-retest reliability was good, with ICCs mostly ranging from moderate to high (ICCs = 0.37-0.93, P < 0.001-0.130), and sensitivity to change was excellent (ηp2 = 0.16-0.81, P < 0.001-0.044). Completion rates for the initial challenge levels were 100%, mean completion time 36.3 minutes, and no clinical events or safety problems were observed. Conclusion: Study results documented on average good validity, test-retest reliability and feasibility, with an extraordinary high responsiveness of the presented game-based assessment in older adults with moderate functional limitations. The innovative, data-based assessment validated in this study may serve as a blueprint for future, tailored assessments for exergaming.

The Industrial Organization of the Syrian Civil War

The Syrian Civil War represents an extreme outlier in terms of the number of insurgent groups which have been engaged in the fighting. These groups have also been remarkably persistent over time, partly due to the fact that rebel in-fighting has been relatively contained. They have also targeted civilians far less than the Syrian Army. These stylized facts run counter to much of the existing literature on multi-party civil wars, which has emphasized the influence of the balance of power on group dynamics. In this article we instead draw upon balance of threat theory, along with insights from the economics of industrial organization, to understand insurgent behavior in the Syrian Civil War, based on a newly compiled dataset of rebel violence. Our research suggests that conflict scholars need to account for factors beyond the balance of power if they are to adequately explain inter-rebel dynamics.

Use of MODIS NDVI Products to Map Tree Mortality Levels in Forests Affected by Mountain Pine Beetle Outbreaks

Extensive bark beetle outbreaks have recently occurred in western North American forests, resulting in overstory tree mortality across millions of hectares. Annual aerial surveys are currently used to operationally monitor bark beetle induced tree mortality, though this method is subjective and can exclude some forest areas. Daily Moderate Resolution Imaging Spectroradiometer (MODIS) satellite data offer a potential alternative means to develop regional tree mortality maps. Accurate methods using such data could aid natural resource managers in surveys of forests with frequent overstory mortality, helping to prioritize forest treatment and restoration activities. This paper discusses a study to test the potential of using MODIS data to detect tree mortality. We developed and tested an approach to use 250-m resolution MODIS Normalized Difference Vegetation Index (NDVI) data products collected during a mountain pine beetle (MPB) outbreak and related tree mortality event in the northern Rocky Mountains of Colorado, USA. The 94 km2 study area is predominantly lodgepole pine forest with most of the MPB-caused mortality occurring between 2003 and 2008. We used a 2.4-m forest conditions map from 2008 aerial multispectral imagery to calculate percentage of mortality within 240-m pixels for use as reference data. Using either daily or 16-day products, MODIS NDVI change products were calculated for 2008 versus either 2000 or 2003 baselines. MODIS change products were used as predictors in linear regression analysis to assess correlation between MODIS data and the aerial percent forest mortality map. Depending on the MODIS product, linear regression analyses yielded r2 values ranging from 0.362 to 0.544 without outliers removed and from 0.406 to 0.570 with extreme outliers removed. Daily MODIS NDVI products from 2003 and 2008 were used with exponential regression to improve the r2 to 0.593. The project showed some MODIS NDVI data potential for mapping percent tree mortality in forests subjected to regional bark beetle outbreaks and severe drought.

An Augmented Lagrangian Approach for Distributed Robust Estimation in Large-Scale Systems

Nonlinear estimation using maximum-likelihood estimation and maximum a posterior probability approaches is frequently employed for large-scale cyber-physical and complex systems in the big data era. Efficient distributed processing and robust estimation algorithms resilient to outliers are of great importance. This paper proposes a novel method for distributed solution of these robust estimation problems with equality constraints based on the augmented Lagrangian method (ALM). Specifically, a novel covariance normalization method and an automatic method for selecting regularization parameter with improved performance are proposed. Under the ALM framework, nonlinear equality constraints and nonsmooth L 1 regularization can be incorporated. The proposed method is illustrated with two emerging applications respectively in robust distributed power system state estimation (DSSE) with nonlinear zero injection constraints and gene regulatory network (GRN) identification. Experimental results in DSSE show that the covariance normalization method improves considerably the convergence speed over the alternating direction method of multipliers algorithm and the robust statistics employed effectively mitigates the adverse effects of extreme outliers. Zero-injection constraints can be effectively incorporated. For GRN identification, putative genes and connectivity for a yeast dataset with 1.57 million variables can be identified via sparsity and piecewise temporal continuity penalties and they generally align well with literature.

A Comparison of Partisan-Gerrymandering Measures

We compare and contrast 14 measures that have been proposed for the purpose of quantifying partisan gerrymandering. We consider measures that, rather than examining the shapes of districts, utilize only the partisan vote distribution among districts. The measures considered are two versions of partisan bias, the efficiency gap and several of its variants, the mean-median difference and the equal vote weight standard, the declination and one variant, and the lopsided-means test. Our primary means of evaluating these measures is a suite of hypothetical elections we classify from the start as fair or unfair. We conclude that the declination is the most successful measure in terms of avoiding false positives and false negatives on the elections considered. We include in a supplementary appendix the most extreme outliers for each measure among historical congressional and state legislative elections.

Directional bivariate quantiles: a robust approach based on the cumulative distribution function

The definition of multivariate quantiles has gained considerable attention in previous years as a tool for understanding the structure of a multivariate data cloud. Due to the lack of a natural ordering for multivariate data, many approaches have either considered geometric generalisations of univariate quantiles or data depths that measure centrality of data points. Both approaches provide a centre-outward ordering of data points but do no longer possess a relation to the cumulative distribution function of the data generating process and corresponding tail probabilities. We propose a new notion of bivariate quantiles that is based on inverting the bivariate cumulative distribution function and therefore provides a directional measure of extremeness as defined by the contour lines of the cumulative distribution function which define the quantile curves of interest. To determine unique solutions, we transform the bivariate data to the unit square. This allows us to introduce directions along which the quantiles are unique. Choosing a suitable transformation also ensures that the resulting quantiles are equivariant under monotonically increasing transformations. We study the resulting notion of bivariate quantiles in detail, with respect to computation based on linear programming and theoretical properties including asymptotic behaviour and robustness. It turns out that our approach is especially useful for data situations that deviate from the elliptical shape typical for ‘normal-like’ bivariate distributions. Moreover, the bivariate quantiles inherit the robustness of univariate quantiles even in case of extreme outliers.

Semi-automated detection of landslide timing using harmonic modelling of satellite imagery, Buckinghorse River, Canada

We manually detected and mapped 66 landslides from Landsat imagery over a 33-year period from 1985 to 2017 in the Buckinghorse River region, British Columbia, Canada. We semi-automatically determined landslide timing using the cumulative difference (CD) between the normalized difference vegetation index (NDVI) and a fitted harmonic sinusoidal curve (CDNDVI). The semi-automated dating method was capable of determining the timing of 80% of the landslides using CDNDVI and 85% of the landslides after detrending CDNDVI (dCDNDVI). The CDNDVI method generally detects landslides too early and the dCDNDVI method is generally too late. Mean absolute errors (in days) are lower for the dCDNDVI (208 and 188) than the CDNDVI (227 and 267), respectively. This study, however, has many examples of extreme outliers with very large errors (>1000 days). Our method is portable to other remote regions as long as vegetation anomalies can be used as an indicator for landslide activity. We conclude that the timeseries of images available in the Landsat Archive are useful for landslide mapping, but the pixel size limits the size of the landslides that can be mapped.

ORELM: A Novel Machine Learning Approach for Prediction of Flyrock in Mine Blasting

Blast-induced flyrock is a hazardous and undesirable phenomenon that may occur in surface mines, especially when blasting takes place near residential areas. Therefore, accurate prediction of flyrock distance is of high significance in the determination of the statutory danger area. To this end, there is a practical need to propose an accurate model to predict flyrock. Aiming at this topic, this study presents two machine learning models, including extreme learning machine (ELM) and outlier robust ELM (ORELM), for predicting flyrock. To the best of our knowledge, this is the first work that investigates the use of ORELM model in the field of flyrock prediction. To construct and verify the proposed ELM and ORELM models, a database including 82 datasets has been collected from the three granite quarry sites in Malaysia. Additionally, artificial neural network (ANN) and multiple regression models were used for comparison. According to the results, both ELM and ORELM models performed satisfactorily, and their performances were far better compared to the performances of ANN and multiple regression models.

An Adaptive Regression Feature Selection Method for Datasets with Outliers

An Adaptive Regression Feature Selection Method for Datasets with Outliers

Visual complexity modelling based on image features fusion of multiple kernels.

Humans’ perception of visual complexity is often regarded as one of the key principles of aesthetic order, and is intimately related to the physiological, neurological and, possibly, psychological characteristics of the human mind. For these reasons, creating accurate computational models of visual complexity is a demanding task. Building upon on previous work in the field (Forsythe et al., 2011; Machado et al., 2015) we explore the use of Machine Learning techniques to create computational models of visual complexity. For that purpose, we use a dataset composed of 800 visual stimuli divided into five categories, describing each stimulus by 329 features based on edge detection, compression error and Zipf’s law. In an initial stage, a comparative analysis of representative state-of-the-art Machine Learning approaches is performed. Subsequently, we conduct an exhaustive outlier analysis. We analyze the impact of removing the extreme outliers, concluding that Feature Selection Multiple Kernel Learning obtains the best results, yielding an average correlation to humans’ perception of complexity of 0.71 with only twenty-two features. These results outperform the current state-of-the-art, showing the potential of this technique for regression.

PENGARUH MURABAHAH, MUSYARAKAH, MUDHARABAH, DAN IJARAH TERHADAP PROFITABILITAS BANK UMUM SYARIAH DI INDONESIA

The aim of this study was to examine the effect of Murabaha, Musyaraka, Mudharaba, and Ijara on Profitability on all Sharia General Bank listed in Otoritas Jasa Keuangan (OJK) period June 2015 – June 2018. The sampling in this study of 13 banks listed in Otoritas Jasa Keuangan period June 2015 – June 2018 with purposive sampling method. Data obtained from the company’s financial statements that have been publication. The number of samples was obtained from 13 companies with total 91 sample. After eliminating the outlier data, which eventually resulted 66 observations that ready to be analyzed and tested. The analysis technique that used in this study is multiple linear regression with a significance level of 5%. The results in this study indicates that (1) Murabaha has significant effect on Profitability, (2) Musyaraka has significant effect on Profitability, (3) Mudharaba has significant effect on Profitability, and (4) Ijara has not significant effect on profitability. Ijara was the only variable in this research that has not significant effect on profitability because the proportion of ijara is the smallest financing of all.

Ensemble of evolving optimal granular experts, OWA aggregation, and time series prediction

This paper presents an online-learning ensemble framework for nonstationary time series prediction. Optimal granular fuzzy rule-based models with different objective functions and constraints are evolved from data streams. Evolving optimal granular systems (eOGS) consider multiobjective optimization, the specificity of information, model compactness, and variability and coverage of the data within the process of modeling data streams. Forecasts of individual base eOGS models are combined using averaging aggregation functions: ordered weighted averaging (OWA), weighted arithmetic mean, median, and linear non-inclusive centered OWA. Some aggregation functions use specific weights for the relevance of the base models and exclude extreme values and outliers. The weights of other aggregation functions are adapted over time based on a quadratic programming problem and the data within a sliding window. This paper investigates whether an online-learning ensemble can outperform individual eOGS models, and which aggregation function provides the most accurate forecasts. Real multivariate weather time series, particularly time series of daily mean temperature, air humidity, and wind speed from different weather stations, such as Paris–Orly, Frankfurt–Main, Reykjavik, and Oslo–Blindern, are used for evaluation. The results show that ensemble schemes outperform individual models. The proposed linear non-inclusive centered OWA function provided the most accurate numerical predictions.

Using adaptively weighted large margin classifiers for robust sufficient dimension reduction

ABSTRACTIn this paper, we combine adaptively weighted large margin classifiers with Support Vector Machine (SVM)-based dimension reduction methods to create dimension reduction methods robust to the presence of extreme outliers. We discuss estimation and asymptotic properties of the algorithm. The good performance of the new algorithm is demonstrated through simulations and real data analysis.

Length of stay for acute myocardial infarction patients in eastern rural area in China from 2001 to 2011

Objective To describe the average hospital length of stay (LOS) and its trend and variation in rural areas of eastern China, and to analyze the hospital characteristics associated with LOS. Methods We collected the data of representative in-patients with acute myocardial infarction (AMI) in eastern rural areas in 2001, 2006 and 2011 by random sampling. The hospital length of stay and its changing trend were described. Hospitals were divided into tertiles-high, middle, low-based on the median LOS; the characteristics of patients, AMI treatment, as well as outcome were compared across the tertiles of hospitals. The association between the hospital characteristics and LOS was analyzed after adjusting for patient-level characteristics. Results We examined the medical records of 2 361 patients hospitalized with AMI from 32 hospitals in the eastern rural areas. The average LOS were (14.7±10)days in 2001, (12.4±10.1)days in 2006, and (11.9±6.5)days in 2011, decreasing annually (F=15.41, P<0.001). When looking at the cohort on the three time points (2001, 2006, and 2011) with ruling out the factors of complication during hospitalization and the LOS outliers, the average LOS still indicated an trend of decline (F=5.44, P=0.02). However, after adjusting for the differences in patient factors, few marked differences in LOS were found over the years (P=0.89). Across the three tertiles of hospitals, categorized as high, middle and low based on the average LOS, a statistical difference existed in the use rate of statin during hospitalization (P<0.05). After adjusting for patient-level characteristics, the hospital characteristics were not significantly associated with the average LOS. Conclusions This study showed a trend of an annual decrease of LOS in patients hospitalized with AMI from the eastern rural areas in China. There is nevertheless still significant room for further improvement. Therefore, it is crucial to effectively advance the standardization for AMI patients of diagnosis and treatment, improve healthcare quality, and reduce patient LOS, which is the very key to enhancing the AMI treatments in the rural region of Eastern China. Key words: Myocardial infarction; Length of stay; Quality of health care; Hospitals, rural