A Lasso-LTS Method for DNA Sequence Classification Based on Beta Wavelet Networks

Wavelet Neural Network (WNN) is attracting interest in field of classification system, because they are universal approximations, particularly due to rapid and accurate representation of nonlinear dynamic systems. The satisfying performance of the WNN depends on an appropriate determination of the Wavelet Neural Network structure. In this paper we provide a new method to solve this problem based on the Least Absolute Shrinkage and Selection Operator (LASSO). At first, the scale of WNN is managed by using the time-frequency locality of wavelet. Furthermore, the unconstrained optimization problem (LASSO) is used to solve the structure and learning of the WNN. This optimization problem can be solved efficiently using the iteratively reweighted least squares (IRLS) and the Least Trimmed Square (LTS) methods to enhance the ineffectiveness; they are applied to train the wavelet neural network. The advantage of the method lies in the oracle properly of the LASSO can guarantee the optimal structure of the WNN. The proposed method has been able to optimize the wavelet neural network and this method is able to classify the DNA sequences. Our goal is to construct predictive models that are highly accurate. In fact, the proposed method permits to avoid the complex problem of form and structure in different clusters of organisms. The empirical results and their classification performances are compared with other methods. We compared the WNN-Lasso model with the other five alignment-free models, i.e., k-tuple, DMK, TSM, AMI, and CV, on several large-scale DNA datasets on the DNA classifying application by means of the K-means method. The experimental results have shown that the WNN-Lasso model outperformed the other models in terms of both the classifying results and the running time. Evenly, in this study, we present our approach consists of three phases. The first one, which is called transformation, is composed of two sub steps; binary codification of the DNA sequences and the Signal Processing of the DNA sequences. The second phase step is the approximation; it is empowered by the use of the Multi Library Wavelet Neural Networks (MLWNN). Finally, the third section, which is the classification of the DNA sequences, is realized by applying the algorithm of k-means classification.