In the present work we extend the traditional C4.5 decision tree method for regression and forecasting of multivariate time series. In the proposed method, time series data is first fuzzified leading to a fuzzy time series (FTS) representation of the data. A fuzzy decision tree (FDT) based on C4.5 is employed to form the knowledge base of the FTS model. The method can deal with high-order and multivariate fuzzy time series, offering an explainable model. The FDT-FTS method is tested with data from IBOVESPA stock market index, which tracks the performance of around 50 most liquid stocks traded on the Sao Paulo Stock Exchange in Brazil. The method is applied to the IBOVESPA mini future contract time series in order to forecast future values using a mix of historical values and technical analysis indicators. This method is compared with Support Vector Regression (SVR) and Random Forest Regression (RFR), both methods implemented in the Scikit-Learn open-source library. The FDT-FTS model was implemented in Python programming language in the open-source pyFTS library. Although all three methods have similar performance, according to the MAPE, SMAPE, RMSE, NRMSE and MAE metrics, the proposed method is computationally faster and explainable.

This paper reports how OptBees, an algorithm inspired by the collective decision-making of bee colonies, performed in the test bed developed for the Special Session & Competition on Real-Parameter Single Objective Optimization at CEC-2014. The test bed includes 30 scalable functions, many of which are both non-separable and highly multi-modal. Results include OptBees' performance on the 10, 30, 50 and 100-dimensional versions of each function.

This paper presents an evolving neuro-fuzzy network approach (eNFN) to model a twin rotor MIMO system (TRMS) with two degrees of freedom in real-time. The TRMS is a fast, nonlinear, open loop unstable time-varying dynamic system, with cross coupling between the rotors. Modeling and control of TRMS require high sampling rates, typically in the order of milliseconds. Actual laboratory implementation shows that eNFN is fast, effective, and accurately models the TRMS in real-time. The eNFN captures the TRMS system dynamics quickly, and develops precise low cost models from the point of view of time and space complexity. The results suggest eNFN as a potential candidate to model complex, fast time-varying dynamic systems in real-time.

This paper presents the OptBees, an optimization algorithm inspired by the processes of collective decision-making by bee colonies. The algorithm was designed with the objective of generating and maintaining diversity, promoting a multimodal search and obtaining multiple local optima without losing the ability of global optimization, thus representing an innovation compared with existent bee-inspired algorithms. It has been tested in five of the twenty-five minimization problems proposed for the Optimization Competition of Real Parameters of the CEC 2005 Special Session on Real-Parameter Optimization, held in the 2005 IEEE Congress on Evolutionary Computation (CEC). The results obtained suggest the suitability of the algorithm to exploit multimodality of problems, being successful in the generation and maintenance of diversity and achieving good quality results in global optimization.

The 2012 FUZZ-IEEE conference competition “Learning Fuzzy Systems from Data” aims to establish the empirical accuracy of fuzzy forecasting algorithms in the domain of prediction of the sales volume of petroleum products. Currently, there are no guidelines or consensus on a best practice methodology. This paper proposes evolving fuzzy linear regression trees (eFT) to extract from both, daily prices and past sales volume data, information of interest to attain accurate forecasts of the next day sales. Essentially, eFT attempts to find spatio-temporal correlations from a historical perspective of competitors' prices and previous sales. A dimension reduction method based on the least angle regression (LARS) algorithm is considered for input variable selection. Computational experiments show that the eFT predictor using LARS is an effective approach to nonlinear time series forecasting providing encouraging results in the competition scenario.

This paper introduces a neural fuzzy network approach for evolving system modeling. The approach uses neofuzzy neurons and a neural fuzzy structure monished with an incremental learning algorithm that includes adaptive feature selection. The feature selection mechanism starts considering one or more input variables from a given set of variables, and decides if a new variable should be added, or if an existing variable should be excluded or kept as an input. The decision process uses statistical tests and information about the current model performance. The incremental learning scheme simultaneously selects the input variables and updates the neural network weights. The weights are adjusted using a gradient-based scheme with optimal learning rate. The performance of the models obtained with the neural fuzzy modeling approach is evaluated considering weather temperature forecasting problems. Computational results show that the approach is competitive with alternatives reported in the literature, especially in on-line modeling situations where processing time and learning are critical.

This paper introduces a class of evolving fuzzy rule-based system as an approach for multivariable Gaussian adaptive fuzzy modeling. The system is an evolving Takagi-Sugeno (eTS) functional fuzzy model, whose rule base can be continuously updated using a new recursive clustering algorithm based on participatory learning. The fuzzy sets of the rule antecedents are multivariable Gaussian membership functions, which have been adopted to preserve information between input variable interactions. The parameters of the membership functions are estimated by the clustering algorithm. A weighted recursive least-squares algorithm updates the parameters of the rule consequents. Experiments considering time-series forecasting and nonlinear system identification are performed to evaluate the performance of the approach proposed. The multivariable Gaussian evolving fuzzy models are compared with alternative evolving fuzzy models and classic models with fixed structures. The results suggest that multivariable Gaussian evolving fuzzy modeling is a promising approach for adaptive system modeling.

This paper introduces a new approach for evolving fuzzy modeling based on a tree structure. The system is a fuzzy linear regression tree whose topology can be continuously updated using a statistical model selection test. A fuzzy linear regression tree is a fuzzy tree with linear model in each leaf. The evolving linear regression approach is evaluated on a forecasting problem and its performance compared against alternative evolving fuzzy models and classic models with fixed structures. The results suggest that evolving fuzzy regression tree is a promising approach for adaptive system modeling.

The achievement of approximation models may constitute a complex computational task, in the cases of models with non-linear relation between parameters and data. This problem becomes even harder when the system to be modeled is subject to noisy data, since the simple minimization of error over a training data set can give rise to misleading models that fit both the system structure and the noise (the phenomenon of modeloverfit). This paper proposes a multiobjective genetic algorithm for guiding the training of ANFIS fuzzy networks. This algorithm considers the complexity of network jointly with the error over the training set as relevant objectives, that should be minimized. Results obtained in three regression problems are presented to show the generalization capacity of models constructed with the proposed methodology.

This paper presents a novel approach for dealing with the structural risk minimization (SRM) applied to a general setting of the machine learning problem. The formulation is based on the fundamental concept that supervised learning is a bi-objective optimization problem in which two conflicting objectives should be minimized. The objectives are related to the empirical training error and the machine complexity. In this paper, one general Q-norm method to compute the machine complexity is presented, and, as a particular practical case, the minimum gradient method (MGM) is derived relying on the definition of the fat-shattering dimension. A practical mechanism for parallel layer perceptron (PLP) network training, involving only quasi-convex functions, is generated using the aforementioned definitions. Experimental results on 15 different benchmarks are presented, which show the potential of the proposed ideas.

This paper addresses the problem of efficiently computing robot navigation functions. Navigation functions are potential functions free of spurious local minima that present an exact solution to the robot motion planning and control problem. Although some methodologies were found in the literature, none of them are easy to implement and generalize for complex shaped workspaces and robots. We discuss some of the difficulties encountered in the current methodologies and propose a novel approach using a Finite Element method for potential field computation.

This work presents a Neo-Fuzzy-Neuron algorithm for the identification of nonlinear dynamic systems at the point of view of a rotor flux observer. The algorithm training is on-line, has low computational cost, does not require previous training and its convergence in one step is proved. The gradient descent method is used for its weights adjustment. Simulation and experimental results demonstrate the effectiveness of the algorithm for flux observer of induction motor drive system.