Selection–fusion approach for classification of datasets with missing values

Ghannad-Rezaie, Mostafa; Soltanian-Zadeh, Hamid; Ying, Hao; Dong, Ming

doi:10.1016/j.patcog.2009.12.003

Cited by 33 publications

(20 citation statements)

References 24 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Despite the innovation regarding the use of genetic algorithm to perform data imputation itself, this approach shows some weaknesses considering scalability and accuracy, since the individuals are codified into a m size vector, instead of using subsets/partitions [2,12]. A modified version of these approaches was proposed for multivariate data [8] and it is important to highlight that this method falls into completecase analysis, since the statistic information used in fitness function are extracted from examples without missing values, thus, important information is lost.…”

Section: Evolutionary Approachesmentioning

confidence: 99%

An Evolutionary Missing Data Imputation Method for Pattern Classification

Lobato

Tadaiesky

Araújo

et al. 2015

Proceedings of the Companion Publication of the 2015 Annual Conference on Genetic and Evolutionary Computation

View full text Add to dashboard Cite

Data analysis plays an important role in our InformationEra; however, most of statistical and machine learning algorithms were not developed to tackle the ubiquitous issue of missing values. In pattern classification, several strategies have been proposed to handle this problem, where missing data imputation is the most used one, which can be viewed as an optimization problem where the goal is to reduce the bias imposed by the absence of information. Although most imputation methods are restricted to one type of variable only (categorical or numerical), they usually ignore information within incomplete instances. To fill these gaps, we propose an evolutionary missing data imputation method for pattern classification, based on a genetic algorithm, which is suitable for mixed-attribute datasets and takes into account information from incomplete instances and model building -more specifically, the classification accuracy. To assess the performance of our method, we used three algorithms in order to represent the three groups of classification methods: 1) rule induction learning, 2) approximate models and 3) lazy learning. Experiments have shown that the proposed method outperforms some well-established missing value treatment methods.

show abstract

Section: Evolutionary Approachesmentioning

confidence: 99%

An Evolutionary Missing Data Imputation Method for Pattern Classification

Lobato

Tadaiesky

Araújo

et al. 2015

Proceedings of the Companion Publication of the 2015 Annual Conference on Genetic and Evolutionary Computation

View full text Add to dashboard Cite

show abstract

“…However, due to the missing values in the feature matrix, feature and sample selections can not be performed directly. We thus partition the feature matrix, together with the target output matrix, into submatrices with only complete data (Ghannad-Rezaie et al, 2010), so that a 2-step multi-task learning algorithm (Obozinski et al, 2006; Zhang and Shen, 2012) can be applied to these sub-matrices to obtain a set of discriminative features and samples. The selected features and samples then form a shrunk, but still incomplete, matrix which is more “friendly” to imputation algorithms, as redundant/noisy features and samples have been removed and there are now a smaller number of missing values that need to be imputed.…”

Section: Introductionmentioning

confidence: 99%

Neurodegenerative disease diagnosis using incomplete multi-modality data via matrix shrinkage and completion

et al. 2014

View full text Add to dashboard Cite

In this work, we are interested in predicting the diagnostic statuses of potentially neurodegenerated patients using feature values derived from multi-modality neuroimaging data and biological data, which might be incomplete. Collecting the feature values into a matrix, with each row containing a feature vector of a sample, we propose a framework to predict the corresponding associated multiple target outputs (e.g., diagnosis label and clinical scores) from this feature matrix by performing matrix shrinkage following by matrix completion. Specifically, we first combine the feature and target output matrices into a large matrix and then partition this large incomplete matrix into smaller submatrices, each consisting of samples with complete feature values (corresponding to a certain combination of modalities) and target outputs. Treating each target output as the outcome of a prediction task, we apply a 2-step multi-task learning algorithm to select the most discriminative features and samples in each submatrix. Features and samples that are not selected in any of the submatrices are discarded, resulting in a shrunk version of the original large matrix. The missing feature values and unknown target outputs of the shrunk matrix is then completed simultaneously. Experimental results using the ADNI dataset indicate that our proposed framework achieves higher classification accuracy at a greater speed when compared with conventional imputation-based classification methods and also yields competitive performance when compared with the state-of-the-art methods.

show abstract

“…The underlying idea of ensemble learning for classification problems is to build a number of base classifiers and then combine their outputs using a fusion rule. It has been shown that classifier ensembles outperform single classifiers for a wide range of classification problems [27][28][29]. The reason is that a combination of multiple classifiers reduces risks associated with choosing an insufficient single classifier.…”

Section: Classifier Ensemblementioning

confidence: 99%