Voxelwise detection of cerebral microbleed in CADASIL patients by leaky rectified linear unit and early stopping

It is important to detect cerebral microbleed voxels from the brain image of cerebral autosomal-dominant arteriopathy with subcortical infarcts and Leukoencephalopathy (CADASIL) patients. Traditional manual method suffers from intra-observe and inter-observe variability. In this study, we used the susceptibility weighted imaging (SWI) to scan 10 CADASIL patients and 10 healthy controls. We used slicing neighborhood processing (SNP) to extract “input” and “target” dataset from the 20 brain volumetric images. Afterwards, the undersampling technique was employed to handle the class-imbalanced problem. The single-hidden layer feedforward neural-network with scaled conjugate gradient was used as the classifier. We compared three activation functions: logistic sigmoid (LOSI), rectified linear unit (ReLU), and leaky rectified linear unit (LReLU). Early stopping and K-fold cross validation (CV) was used to avoid overfitting and statistical analysis. In the experiment, we generated 68,847 CMB voxels, and 68,829 non-CMB voxels. We observed that LReLU achieved the best result with a sensitivity of 93.05%, a specificity of 93.06%, and an accuracy of 93.06%. We also observed the effect of early stopping and K-fold CV. We found the optimal number of hidden neuron was 10 by grid searching method. Besides, our method performs better than three state-of-the-art methods. The results show our method is promising. In addition, LReLU is a better activation function that may replace traditional logistic sigmoid function in other applications.     

Multi-armed bandit heterogeneous ensemble learning for imbalanced data

One of the most widely used approaches to the class-imbalanced issue is ensemble learning. The base classifier is trained using an unbalanced training set in the conventional ensemble learning approach. We are unable to select the best suitable resampling method or base classifier for the training set, despite the fact that researchers have examined employing resampling strategies to balance the training set. A multi-armed bandit heterogeneous ensemble framework was developed as a solution to these issues. This framework employs the multi-armed bandit technique to pick the best base classifier and resampling techniques to build a heterogeneous ensemble model. To obtain training sets, we first employ the bagging technique. Then, we use the instances from the out-of-bag set as the validation set. In general, we consider the basic classifier combination with the highest validation set score to be the best model on the bagging subset and add it to the pool of model. The classification performance of the multi-armed bandit heterogeneous ensemble model is then assessed using 30 real-world imbalanced data sets that were gathered from UCI, KEEL, and HDDT. The experimental results demonstrate that, under the two assessment metrics of AUC and Kappa, the proposed heterogeneous ensemble model performs competitively with other nine state-of-the-art ensemble learning methods. At the same time, the findings of the experiment are confirmed by the statistical findings of the Friedman test and Holm's post-hoc test.     

LCMine: An efficient algorithm for mining discriminative regularities and its application in supervised classification

In this paper, we introduce an efficient algorithm for mining discriminative regularities on databases with mixed and incomplete data. Unlike previous methods, our algorithm does not apply an a priori discretization on numerical features; it extracts regularities from a set of diverse decision trees, induced with a special procedure. Experimental results show that a classifier based on the regularities obtained by our algorithm attains higher classification accuracy, using fewer discriminative regularities than those obtained by previous pattern-based classifiers. Additionally, we show that our classifier is competitive with traditional and state-of-the-art classifiers.     

软件缺陷集成预测模型研究

利用单一分类器构造的缺陷预测模型已经遇到了性能瓶颈, 而集成分类器相比单一分类器往往具有显著的性能优势。以构造高效的集成缺陷预测模型为出发点, 比较了七种不同类型集成分类器的算法和特点。在14个基准数据集上的实验显示, 部分集成预测模型的性能优于基于朴素贝叶斯的单一预测模型。其中, 基于投票的集成分类框架具有最优的预测性能以及统计学意义上的性能优势显著性, 随机森林算法次之。Stacking集成框架也具有较强的泛化能力。     

Combining Classifiers with Meta Decision Trees

The paper introduces meta decision trees (MDTs), a novel method for combining multiple classifiers. Instead of giving a prediction, MDT leaves specify which classifier should be used to obtain a prediction. We present an algorithm for learning MDTs based on the C4.5 algorithm for learning ordinary decision trees (ODTs). An extensive experimental evaluation of the new algorithm is performed on twenty-one data sets, combining classifiers generated by five learning algorithms: two algorithms for learning decision trees, a rule learning algorithm, a nearest neighbor algorithm and a naive Bayes algorithm. In terms of performance, stacking with MDTs combines classifiers better than voting and stacking with ODTs. In addition, the MDTs are much more concise than the ODTs and are thus a step towards comprehensible combination of multiple classifiers. MDTs also perform better than several other approaches to stacking.     

On the use of ROC analysis for the optimization of abstaining classifiers

Classifiers that refrain from classification in certain cases can significantly reduce the misclassification cost. However, the parameters for such abstaining classifiers are often set in a rather ad-hoc manner. We propose a method to optimally build a specific type of abstaining binary classifiers using ROC analysis. These classifiers are built based on optimization criteria in the following three models: cost-based, bounded-abstention and bounded-improvement. We show that selecting the optimal classifier in the first model is similar to known iso-performance lines and uses only the slopes of ROC curves, whereas selecting the optimal classifier in the remaining two models is not straightforward. We investigate the properties of the convex-down ROCCH (ROC Convex Hull) and present a simple and efficient algorithm for finding the optimal classifier in these models, namely, the bounded-abstention and bounded-improvement models. We demonstrate the application of these models to effectively reduce misclassification cost in real-life classification systems. The method has been validated with an ROC building algorithm and cross-validation on 15 UCI KDD datasets. An early version of this paper was published at ICML2005. Action Editor: Johannes Fürnkranz.     

Neural network ensembles: combining multiple models for enhanced performance using a multistage approach

Abstract: Neural network ensembles (sometimes referred to as committees or classifier ensembles) are effective techniques to improve the generalization of a neural network system. Combining a set of neural network classifiers whose error distributions are diverse can generate better results than any single classifier. In this paper, some methods for creating ensembles are reviewed, including the following approaches: methods of selecting diverse training data from the original source data set, constructing different neural network models, selecting ensemble nets from ensemble candidates and combining ensemble members' results. In addition, new results on ensemble combination methods are reported.     

Troika - An improved stacking schema for classification tasks

Stacking is a general ensemble method in which a number of base classifiers are combined using one meta-classifier which learns their outputs. Such an approach provides certain advantages: simplicity; performance that is similar to the best classifier; and the capability of combining classifiers induced by different inducers. The disadvantage of stacking is that on multiclass problems, stacking seems to perform worse than other meta-learning approaches. In this paper we present Troika, a new stacking method for improving ensemble classifiers. The new scheme is built from three layers of combining classifiers. The new method was tested on various datasets and the results indicate the superiority of the proposed method to other legacy ensemble schemes, Stacking and StackingC, especially when the classification task consists of more than two classes.     

Cascade Generalization

Using multiple classifiers for increasing learning accuracy is an active research area. In this paper we present two related methods for merging classifiers. The first method, Cascade Generalization, couples classifiers loosely. It belongs to the family of stacking algorithms. The basic idea of Cascade Generalization is to use sequentially the set of classifiers, at each step performing an extension of the original data by the insertion of new attributes. The new attributes are derived from the probability class distribution given by a base classifier. This constructive step extends the representational language for the high level classifiers, relaxing their bias. The second method exploits tight coupling of classifiers, by applying Cascade Generalization locally. At each iteration of a divide and conquer algorithm, a reconstruction of the instance space occurs by the addition of new attributes. Each new attribute represents the probability that an example belongs to a class given by a base classifier. We have implemented three Local Generalization Algorithms. The first merges a linear discriminant with a decision tree, the second merges a naive Bayes with a decision tree, and the third merges a linear discriminant and a naive Bayes with a decision tree. All the algorithms show an increase of performance, when compared with the corresponding single models. Cascade also outperforms other methods for combining classifiers, like Stacked Generalization, and competes well against Boosting at statistically significant confidence levels.     

一种基于改进的支持向量机多分类器图像标注方法

针对多标签图像标注问题,提出一种改进的支持向量机多分类器图像标注方法。首先引入直方图交叉距离作为核函数,然后把传统支持向量机的输出值变换为样本到超平面的距离。基于这两点改进,采用一种特征选择方法,从众多的图像特征中,选择那些相互之间冗余度较小的视觉特征,分别建立分类器,最终形成以距离大小为判别依据的支持向量机多分类器模型。此外,在建立分类器时,考虑到训练图像中不同标签类样本分布的不均匀,引入了一个关于图像类标签的概率分布值做为分类器的权重系数。实验采用ImageCLEF提供的图像标注数据集,在其上的实验验证了所采用的特征选择算法和多分类模型的有效性,其标注精度要优于其他传统分类模型,并且,实验结果与最新的方法相比也具有一定的竞争力。