Commenced in January 2007
Frequency: Monthly
Edition: International
Paper Count: 4

imputation Related Abstracts

4 A Review of Methods for Handling Missing Data in the Formof Dropouts in Longitudinal Clinical Trials

Authors: A. Satty, H. Mwambi

Abstract:

Much clinical trials data-based research are characterized by the unavoidable problem of dropout as a result of missing or erroneous values. This paper aims to review some of the various techniques to address the dropout problems in longitudinal clinical trials. The fundamental concepts of the patterns and mechanisms of dropout are discussed. This study presents five general techniques for handling dropout: (1) Deletion methods; (2) Imputation-based methods; (3) Data augmentation methods; (4) Likelihood-based methods; and (5) MNAR-based methods. Under each technique, several methods that are commonly used to deal with dropout are presented, including a review of the existing literature in which we examine the effectiveness of these methods in the analysis of incomplete data. Two application examples are presented to study the potential strengths or weaknesses of some of the methods under certain dropout mechanisms as well as to assess the sensitivity of the modelling assumptions.

Keywords: Sensitivity Analysis, incomplete longitudinal clinical trials, missing at random (MAR), imputation, weighting methods

Procedia PDF Downloads 269
3 Bias-Corrected Estimation Methods for Receiver Operating Characteristic Surface

Authors: Khanh To Duc, Monica Chiogna, Gianfranco Adimari

Abstract:

With three diagnostic categories, assessment of the performance of diagnostic tests is achieved by the analysis of the receiver operating characteristic (ROC) surface, which generalizes the ROC curve for binary diagnostic outcomes. The volume under the ROC surface (VUS) is a summary index usually employed for measuring the overall diagnostic accuracy. When the true disease status can be exactly assessed by means of a gold standard (GS) test, unbiased nonparametric estimators of the ROC surface and VUS are easily obtained. In practice, unfortunately, disease status verification via the GS test could be unavailable for all study subjects, due to the expensiveness or invasiveness of the GS test. Thus, often only a subset of patients undergoes disease verification. Statistical evaluations of diagnostic accuracy based only on data from subjects with verified disease status are typically biased. This bias is known as verification bias. Here, we consider the problem of correcting for verification bias when continuous diagnostic tests for three-class disease status are considered. We assume that selection for disease verification does not depend on disease status, given test results and other observed covariates, i.e., we assume that the true disease status, when missing, is missing at random. Under this assumption, we discuss several solutions for ROC surface analysis based on imputation and re-weighting methods. In particular, verification bias-corrected estimators of the ROC surface and of VUS are proposed, namely, full imputation, mean score imputation, inverse probability weighting and semiparametric efficient estimators. Consistency and asymptotic normality of the proposed estimators are established, and their finite sample behavior is investigated by means of Monte Carlo simulation studies. Two illustrations using real datasets are also given.

Keywords: imputation, missing at random, inverse probability weighting, ROC surface analysis

Procedia PDF Downloads 246
2 Ensemble Methods in Machine Learning: An Algorithmic Approach to Derive Distinctive Behaviors of Criminal Activity Applied to the Poaching Domain

Authors: Zachary Blanks, Solomon Sonya

Abstract:

Poaching presents a serious threat to endangered animal species, environment conservations, and human life. Additionally, some poaching activity has even been linked to supplying funds to support terrorist networks elsewhere around the world. Consequently, agencies dedicated to protecting wildlife habitats have a near intractable task of adequately patrolling an entire area (spanning several thousand kilometers) given limited resources, funds, and personnel at their disposal. Thus, agencies need predictive tools that are both high-performing and easily implementable by the user to help in learning how the significant features (e.g. animal population densities, topography, behavior patterns of the criminals within the area, etc) interact with each other in hopes of abating poaching. This research develops a classification model using machine learning algorithms to aid in forecasting future attacks that is both easy to train and performs well when compared to other models. In this research, we demonstrate how data imputation methods (specifically predictive mean matching, gradient boosting, and random forest multiple imputation) can be applied to analyze data and create significant predictions across a varied data set. Specifically, we apply these methods to improve the accuracy of adopted prediction models (Logistic Regression, Support Vector Machine, etc). Finally, we assess the performance of the model and the accuracy of our data imputation methods by learning on a real-world data set constituting four years of imputed data and testing on one year of non-imputed data. This paper provides three main contributions. First, we extend work done by the Teamcore and CREATE (Center for Risk and Economic Analysis of Terrorism Events) research group at the University of Southern California (USC) working in conjunction with the Department of Homeland Security to apply game theory and machine learning algorithms to develop more efficient ways of reducing poaching. This research introduces ensemble methods (Random Forests and Stochastic Gradient Boosting) and applies it to real-world poaching data gathered from the Ugandan rain forest park rangers. Next, we consider the effect of data imputation on both the performance of various algorithms and the general accuracy of the method itself when applied to a dependent variable where a large number of observations are missing. Third, we provide an alternate approach to predict the probability of observing poaching both by season and by month. The results from this research are very promising. We conclude that by using Stochastic Gradient Boosting to predict observations for non-commercial poaching by season, we are able to produce statistically equivalent results while being orders of magnitude faster in computation time and complexity. Additionally, when predicting potential poaching incidents by individual month vice entire seasons, boosting techniques produce a mean area under the curve increase of approximately 3% relative to previous prediction schedules by entire seasons.

Keywords: Statistical Analysis, Machine Learning, Wildlife Protection, stochastic gradient boosting, imputation, random forests, ensemble methods

Procedia PDF Downloads 154
1 Effect of Genuine Missing Data Imputation on Prediction of Urinary Incontinence

Authors: Suzan Arslanturk, Mohammad-Reza Siadat, Theophilus Ogunyemi, Ananias Diokno

Abstract:

Missing data is a common challenge in statistical analyses of most clinical survey datasets. A variety of methods have been developed to enable analysis of survey data to deal with missing values. Imputation is the most commonly used among the above methods. However, in order to minimize the bias introduced due to imputation, one must choose the right imputation technique and apply it to the correct type of missing data. In this paper, we have identified different types of missing values: missing data due to skip pattern (SPMD), undetermined missing data (UMD), and genuine missing data (GMD) and applied rough set imputation on only the GMD portion of the missing data. We have used rough set imputation to evaluate the effect of such imputation on prediction by generating several simulation datasets based on an existing epidemiological dataset (MESA). To measure how well each dataset lends itself to the prediction model (logistic regression), we have used p-values from the Wald test. To evaluate the accuracy of the prediction, we have considered the width of 95% confidence interval for the probability of incontinence. Both imputed and non-imputed simulation datasets were fit to the prediction model, and they both turned out to be significant (p-value < 0.05). However, the Wald score shows a better fit for the imputed compared to non-imputed datasets (28.7 vs. 23.4). The average confidence interval width was decreased by 10.4% when the imputed dataset was used, meaning higher precision. The results show that using the rough set method for missing data imputation on GMD data improve the predictive capability of the logistic regression. Further studies are required to generalize this conclusion to other clinical survey datasets.

Keywords: rough set, imputation, clinical survey data simulation, genuine missing data, predictive index

Procedia PDF Downloads 24