Search

Imbalanced problem occurs when the size of one class, i.e. the minority class, is much lower than that of the other classes, i.e. the majority classes. Conventional data level methods are employed as the preprocessing approaches to balance the datasets before the classifier learning. Since the balanced data remains unchanged during the learning process, one pre-deleted sample would never be used to train the classifier, which may result in information loss. To solve this problem, this work presents an One-sided Dynamic Undersampling (ODU) technique which adopts all samples in the training process, and dynamically determines whether a majority sample should be used for the classifier learning. Thus, ODU can dynamically undersample the majority class to balance the dataset. To validate the effectiveness of ODU, we integrate it into No-Propagation neural networks to propose an ODU No-Propagation Neural Networks (ODUNPNN). ODUNPNN takes all training samples into consideration, and dynamically undersamples majority class after each iteration, i.e. ODUNPNN integrates undersampling approach into the classifier learning process. Experimental results on both synthetic and real-world imbalance datasets demonstrate that ODUNPNN outperforms the NPNN-based algorithms, and results in comparative performance compared with LASVM-AL, EasyEnsemble, and DyS on real-world imbalance datasets. The contributions of this paper are: (1) ODUNPNN integrates undersampling approach into the classifier learning process. (2) ODUNPNN dynamically balances training data in each iteration. (3) ODU technique can be integrated into other classification learning machines. [ABSTRACT FROM AUTHOR]

Extreme learning machine (ELM), a simple single-hidden-layer feed-forward neural network with fast implementation, has been successfully applied in many fields. This paper proposes an ELM with a constructional structure (CS-ELM) for improving the performance of ELM in dealing with regression problems. In the CS-ELM, there are some partial input subnets (PISs). The first step in designing the PISs is to divide the data-attribute-space into several sub-spaces through using an improved extension clustering algorithm (IECA). The input data attributes in the same sub-space can build a PIS and the similar information of the data attributes is stored in the corresponding PIS. Additionally, a double parallel structure is applied in the CS-ELM, in which there is a special channel that directly connects the input layer neurons to the output layer neurons. In this regard, the proposed procedure can be called ELM with a data-attribute-space-oriented double parallel (DASODP) structure (DASODP–ELM). To test the validity of the proposed method, it is applied to 4 regression applications. The experimental results indicate that, compared with ELM, DASODP–ELM with less number of parameters can achieve higher regression precision in the generalization phase. [ABSTRACT FROM AUTHOR]