Martin Macaš

Robot Path Planning using Particle Swarm Optimization of Ferguson Splines

Robot path planning problem is one of most important task mobile robots. This paper proposes an original approach using a path description by string of cubic splines. Such path is easy executable and natural for car-like robot. Furthermore, it is possible to ensure smooth derivation in connections of particular splines. In this case, the path planning is equivalent to optimization of parameters of splines. An evolutionary technique called particle swarm optimization (PSO) was used hereunder due to its relatively fast convergence and global search character. Various settings of PSO parameters were tested and the best setting was compared to two classical mobile robot path planning algorithms.

Wrapper feature selection for small sample size data driven by complete error estimates

This paper focuses on wrapper-based feature selection for a 1-nearest neighbor classifier. We consider in particular the case of a small sample size with a few hundred instances, which is common in biomedical applications. We propose a technique for calculating the complete bootstrap for a 1-nearest-neighbor classifier (i.e., averaging over all desired test/train partitions of the data). The complete bootstrap and the complete cross-validation error estimate with lower variance are applied as novel selection criteria and are compared with the standard bootstrap and cross-validation in combination with three optimization techniques - sequential forward selection (SFS), binary particle swarm optimization (BPSO) and simplified social impact theory based optimization (SSITO). The experimental comparison based on ten datasets draws the following conclusions: for all three search methods examined here, the complete criteria are a significantly better choice than standard 2-fold cross-validation, 10-fold cross-validation and bootstrap with 50 trials irrespective of the selected output number of iterations. All the complete criterion-based 1NN wrappers with SFS search performed better than the widely-used FILTER and SIMBA methods. We also demonstrate the benefits and properties of our approaches on an important and novel real-world application of automatic detection of the subthalamic nucleus.

Hybridized Swarm Metaheuristics for Evolutionary Random Forest Generation

In many industry and research areas, data mining is a crucial process. This paper presents an evolving structure of classifiers (random forest) where the trees are generated by hybrid method combining ant colony metaheuristics and evolutionary computing technique. The method benefits from the stochastic process and population approach, which allows the algorithm to evolve more efficiently than each method alone. As the method is similar to random forest generation, it can be also used for feature selection. The paper also discusses the parameter estimation for the method. Tests on real data (UCI and real biomedical data) have been performed and evaluated. The average accuracy of the method over MIT-BIH database with normalized data and equalized classes is sensitivity 93.22 % and specificity 87.13 %.

Classification of the emotional states based on the EEG signal processing

The paper proposes a method for the classification of EEG signal based on machine learning methods. We analyzed the data from an EEG experiment consisting of affective picture stimuli presentation, and tested automatic recognition of the individual emotional states from the EEG signal using Bayes classifier. The mean accuracy was about 75 percent, but we were not able to select universal features for classification of all subjects, because of inter-individual differences in the signal. We also identified correlation between the classification error and the extroversion-introversion personality trait measured by EPQ-R test. Introverts have lower excitation threshold so we are able to detect the differences in their EEG activity with better accuracy. Furthermore, the use of Kohonens self-organizing map for visualization is suggested and demonstrated on one subject.

Social impact theory based optimizer

This paper introduces a novel stochastic and population-based binary optimization method inspired by social psychology. It is called Social Impact Theory based Optimization (SITO). The method has been developed with the use of some simple modifications of simulations of Latanes Dynamic Social Impact Theory. The usability of the algorithm is demonstrated via experimental testing on some test problems. The results showed that the initial version of SITO performs comparably to the simple Genetic Algorithm (GA) and the binary Particle Swarm Optimization (bPSO).

Towards biological plausibility of electronic noses

The paper presents a novel encoding scheme for neuronal code generation for odour recognition using an electronic nose (EN). This scheme is based on channel encoding using multiple Gaussian receptive fields superimposed over the temporal EN responses. The encoded data is further applied to a spiking neural network (SNN) for pattern classification. Two forms of SNN, a back-propagation based SpikeProp and a dynamic evolving SNN are used to learn the encoded responses. The effects of information encoding on the performance of SNNs have been investigated. Statistical tests have been performed to determine the contribution of the SNN and the encoding scheme to overall odour discrimination. The approach has been implemented in odour classification of orthodox black tea (Kangra-Himachal Pradesh Region) thereby demonstrating a biomimetic approach for EN data analysis.

Social Impact based Approach to Feature Subset Selection

The interactions taking place in the society could be a source of rich inspiration for the development of novel computational methods. This paper describes an application of two optimization methods based on the idea of social interactions. The first one is the Social Impact Theory based Optimizer - a novel method directly inspired by and based on the Dynamic Theory of Social Impact known from social psychology. The second one is the binary Particle Swarm Optimization - well known optimization technique, which could be understood as to be inspired by decision making process in a group. The two binary optimization methods are applied in the area of automatic pattern classification to selection of an optimal subset of classifier’s inputs. The testing is performed using four datasets from UCI repository. The results show the ability of both methods to significantly reduce input dimensionality and simultaneously keep up the generalization ability.

Innovative default prediction approach

A company default prediction approach based on the magic square area.Comparison of commonly used classifiers with the new proposed method.The diagram shape used for a company assessment.Performance of the proposed method is comparable to other widely used approaches. This paper introduces a new scoring method for company default prediction. The method is based on a modified magic square (a spider diagram with four perpendicular axes) which is used to evaluate economic performance of a country. The evaluation is quantified by the area of a polygon, whose vertices are points lying on the axes. The axes represent economic indicators having significant importance for an economic performance evaluation. The proposed method deals with magic square limitations; e.g. an axis zero point not placed in the axes origins, and extends its usage for an arbitrary (higher than 3) number of variables. This approach is applied on corporations to evaluate their economic performance and identify the companies suspected to default. In general, a company score reflects their economic performance; it is calculated as a polygon area. The proposed method is based on the identification of the parameters (axes order, parameters weights and angles between axes) needed to achieve maximum possible model performance. The developed method uses company financial ratios from its financial statements (debt ratio, return on costs etc.) and the information about a company default or bankruptcy as primary input data. The method is based on obtaining a maximum value of the Gini (or Kolmogorov-Smirnov) index that reflects the quality of the ordering of companies according to their score values. Defaulted companies should have a lower score than non-defaulted companies. The number of parameter groups (axes order, parameters weights and angles between axes) can be reduced without a negative impact on the model performance. Historical data is used to set up model parameters for the prediction of possible future companies default. In addition, the methodology allows calculating the threshold value of the score to separate the companies that are suspicious to the default from other companies. A threshold value is also necessary for a model true positive rate and true negative rate calculations. Training and validation processes for the developed model were performed on two independent and disjunct datasets. The performance of the proposed method is comparable to other methods such as logistic regression and neural networks. One of the major advantages of the proposed method is a graphical interpretation of a company score in the form of a diagram enabling a simple illustration of individual factor contribution to the total score value.

Wrapper Feature Selection Significantly Improves Nonlinear Prediction of Electricity Spot Prices

The paper describes the selection of input delays for Focused Time Delay Neural Network (FTDNN). The problem is understood as a feature subset selection problem, where one looks for a set of features (input delays) that minimizes the mean absolute percentage error. This combinatorial optimization problem is solved using sequential forward search. First, an application of the prediction method to hourly Ontario electricity price forecasting is presented, demonstrating the importance of the feature selection. Although the network with only one hidden unit was used, the wrapper based feature selection caused that it outperforms all state-of the art approaches considered for comparison.

Simplified Social Impact Theory Based Optimizer in Feature Subset Selection

This chapter proposes a simplification of the original Social Impact Theory based Optimizer (oSITO). Based on the experiments with seven benchmark datasets it is shown that the novel method called simplified Social Impact Theory based Optimizer (sSITO) does not degrade the optimization abilities and even leads to smaller testing error and better dimensionality reduction. From these points of view, it also outperforms another well known social optimizer - the binary Particle Swarm Optimization algorithm. The main advantages of the method are the simple implementation and the small number of parameters (two). Additionally, it is empirically shown that the sSITO method even outperforms the nearest neighbor margin based SIMBA algorithm.