International Journal on Artificial Intelligence Tools

A multi-stage krill herd (MSKH) algorithm is presented to fully exploit the global and local search abilities of the standard krill herd (KH) optimization method. The proposed method involves exploration and exploitation stages. The exploration stage uses the basic KH algorithm to select a good candidate solution set. This phase is followed by fine-tuning a good candidate solution in the exploitation stage with a focused local mutation and crossover (LMC) operator in order to enhance the reliability of the method for solving global numerical optimization problems. Moreover, the elitism scheme is introduced into the MSKH method to guarantee the best solution. The performance of MSKH is verified using twenty-five standard and rotated and shifted benchmark problems. The results show the superiority of the proposed algorithm to the standard KH and other well-known optimization methods.

We improve the state-of-the-art method for checking the consistency of large qualitative spatial networks that appear in the Web of Data by exploiting the scale-free-like structure observed in their constraint graphs. We propose an implementation scheme that triangulates the constraint graphs of the input networks and uses a hash table based adjacency list to efficiently represent and reason with them. We generate random scale-free-like qualitative spatial networks using the Barabási-Albert (BA) model with a preferential attachment mechanism. We test our approach on the already existing random datasets that have been extensively used in the literature for evaluating the performance of qualitative spatial reasoners, our own generated random scale-free-like spatial networks, and real spatial datasets that have been made available as Linked Data. The analysis and experimental evaluation of our method presents significant improvements over the state-of-the-art approach, and establishes our implementation as the only possible solution to date to reason with large scale-free-like qualitative spatial networks efficiently.

Case retrieval, case reuse and case retention are critical to the reasoning performance of the traditional case-based reasoning (CBR) model. In this paper, the integrated use of template reduction technology (TR), genetic algorithms (GA), nearest neighbor (NN) rules and group decision-making (GDM) establishes the CBR-GDM model. First, the TR method of the case base is introduced. Then, an attribute weights optimization using GA is discussed in the case retrieval phase. After that, a case reuse method is carried out with NN and GDM. Finally, 10 data sets from UCI are used to carry out a comparison experiment by 5-fold cross-validation. The classification accuracy rate and the classification efficiency are analyzed under the small samples, before and after the data reduction. The results show that, combined with TR, GA and GDM, the pattern classification performance by CBR can be improved.

Intrusion Detection Systems have considerable importance in preventing security threats and protecting computer networks against attackers. So far, various classification approaches using data mining and machine learning techniques have been proposed to the problem of intrusion detection. However, using single classifier systems for intrusion detection suffers from some limitations including lower detection rate for low-frequent attacks, detection instability, and complexity in training process. Ensemble classifier systems combine several individual classifiers and obtain a classifier with higher performance. In this paper, we propose a new ensemble classifier using Radial Basis Function (RBF) neural networks and fuzzy clustering in order to increase detection accuracy and stability, reduce false positives, and provide higher detection rate for low-frequent attacks. We also use a hybrid combination method to aggregate the individual predictions of the base classifiers, which helps to increase detection accuracy. The experimental results on NSL-KDD data set demonstrate that our proposed system has a higher detection accuracy compared to other wellknown classification systems. It also performs more effectively for detection of low-frequent attacks. Furthermore, the proposed ensemble method offers better performance compared to popular ensemble methods.

Data clustering is a common data mining techniques used in many applications such as data analysis and pattern recognition. K-means algorithm is the common clustering method which has fallen into the trap of local optimization and does not always create the optimized response to the problem, although having more advantages such as high speed. Artificial bee colony (ABC) is a novel biological-inspired optimization algorithm, having the advantage of less control parameters, strong global optimization ability and easy to implement. However, there are still some problems in ABC algorithm, like inability to find the best solution from all possible solutions. Due to the large step of searching equation in ABC, the chance of skipping the true solution is high. Therefore, in this paper, to balance the diversity and convergence ability of the ABC, Mantegna Lévy distribution random walk is proposed and incorporated with ABC. The new algorithm, ABCL, brings the power of the Artificial Bee Colony algorithm to the K-means algorithm. The proposed algorithm benefits from Mantegna Lévy distribution to promote the ABC algorithm in solving the number of functional evaluation and also obtaining better convergence speed and high accuracy in a short time. We empirically evaluate the performance of our proposed method on nine standard datasets taken from the UCI Machine Learning Repository. The experimental results show that the proposed algorithm has ability to obtain better results in terms of convergence speed, accuracy, and reducing the number of functional evaluation.

Mode analysis in logic programs has been used mainly for code optimization. The mode analysis in this paper supports the program construction process. It is applied to partially complete logic programs. The program construction process is based on schema refinements and refinements by data type operations. Refinements by data type operations are at the end of the refinement process. This mode analysis supports the proper application of refinements by data type operations. In addition, it checks that the declared modes as defined by the Data Type (DT) operations are consistent with the inferred runtime modes. We have implemented an algorithm for mode analysis based on minimal function graphs. An overview of our logic program development method and the denotational semantics of the analysis framework are presented in this paper.

To reduce the spatial complexity and search time of minimum hitting sets in model-based diagnosis (MBD), a new algorithm for searching minimum hitting sets of MBD is proposed in this paper, which can use the characteristics of minimum hitting sets and the idea of spider prey in biology. We call it cobweb search algorithm. In the algorithm, the generation of visit spiders and search strategy are proposed, and the visit spider that can find the visit path of cobweb to search all minimum hitting sets within a cobweb is constructed. Based on the experiment comparisons, Cobweb search algorithm has better performance than other algorithms of searching minimum hitting sets. As an example, a 14-node power distribution network model is constructed. The diagnosis process with MBD is introduced and analyzed in detail, at the same time the cobweb search algorithm is applied in power distribution network fault diagnosis. The experiment results verify the effectiveness and superiority of the proposed algorithm.

Cluster validation is an essential technique in all cluster applications. Several validation methods measure the accuracy of cluster structure. Typical methods are geometric, where only distance and membership form the core of validation. Yao's decision theory is a novel approach for cluster validation, which evolved loss calculations and probabilistic based measure for determining the cluster quality. Conventional rough set algorithms have utilized this validity measure. This paper propagates decision theory, an unprecedented validation scheme for Rough-Fuzzy clustering by resolving loss and probability calculations to predict the risk measure in clustering techniques. Experiments with synthetic and UCI datasets have been performed, proven to deduce the optimal number of clusters overcoming the downsides of traditional validation frameworks. The proposed index can also be applied to other clustering algorithms and extends the usefulness in business oriented data mining.

This paper intends to propose an integrated hybrid algorithm for training radial basis function neural network (RBFNN) learning. The proposed integrated of particle swarm and genetic algorithm based optimization (IPGO) algorithm is composed of two approaches based on particle swarm optimization (PSO) and genetic algorithm (GA) for gathering both their virtues to improve the learning performance of RBFNN. The diversity of individuals results in higher chance to search in the direction of global optimal instead of being confined to local optimal particularly in problem with higher complexity. The IPGO algorithm with PSO-based and GA-based approaches has shown promising results in some benchmark problems with three continuous test functions. After proposing the algorithm for these problems with result providing its outperforming performance, this paper supplements a practical application case for the papaya milk sales forecasting to expound the superiority of the IPGO algorithm. In addition, model evaluation results of the case have showed that the IPGO algorithm outperforms other algorithms and auto-regressive moving average (ARMA) models in terms of forecasting accuracy and execution time.

Financial time series forecasting has become a challenge because it is noisy, non-stationary and chaotic. To overcome this limitation, this paper uses empirical mode decomposition (EMD) to aid the financial time series forecasting and proposes an approach via combining ARIMA and SVR (Support Vector Regression) to forecast. The approach contains four steps: (1) using ARIMA to analyze the linear part of the original time series; (2) EMD is used to decompose the dynamics of the non-linear part into several intrinsic mode function (IMF) components and one residual component; (3) developing a SVR model using the above IMFs and residual components as inputs to model the nonlinear part; (4) combining the forecasting results of linear model and nonlinear model. To verify the effectiveness of the proposed approach, four stock indices are chosen as the forecasting targets. Comparing with some existing state-of-the-art models, the proposed approach gives superior results.

In this paper we investigate the usefulness of morphosyntactic information as well as clustering in modeling Polish for automatic speech recognition. Polish is an inflectional language, thus we investigate the usefulness of an N-gram model based on morphosyntactic features. We present how individual types of features influence the model and which types of features are best suited for building a language model for automatic speech recognition. We compared the results of applying them with a class-based model that is automatically derived from the training corpus. We show that our approach towards clustering performs significantly better than frequently used SRI LM clustering method. However, this difference is apparent only for smaller corpora.

Differing from the conventional back-propagation (BP) neural networks, a novel multi-input Euler polynomial neural network, in short, MIEPNN (specifically, 4-input Euler polynomial neural network, 4IEPNN) is established and investigated in this paper. In order to achieve satisfactory performance of the established MIEPNN, a weights and structure determination (WASD) algorithm with pruning-while-growing (PWG) and twice-pruning (TP) techniques is built up for the established MIEPNN. By employing the weights direct determination (WDD) method, the WASD algorithm not only determines the optimal connecting weights between hidden layer and output layer directly, but also obtains the optimal number of hidden-layer neurons. Specifically, a sub-optimal structure is obtained via the PWG technique, then the redundant hidden-layer neurons are further pruned via the TP technique. Consequently, the optimal structure of the MIEPNN is obtained. To provide a reasonable choice in practice, several different MATLAB computing routines related to the WDD method are studied. Comparative numerical-experiment results of the 4IEPNN using these different MATLAB computing routines and the standard multi-layer perceptron (MLP) neural network further verify the superior performance and efficacy of the proposed MIEPNN equipped with the WASD algorithm including PWG and TP techniques in terms of training, testing and predicting.