Tadeusz Wieczorek

Comparison of feature ranking methods based on information entropy

A comparison between five feature ranking methods based on entropy is presented on artificial and real datasets. Feature ranking method using /spl chi//sup 2/ statistics gives results that are very similar to the entropy-based methods. The quality of feature rankings obtained by these methods is evaluated using the decision tree and the nearest neighbor classifier with growing number of most important features. Significant differences are found in some cases, but there is no single best index that works best for all data and all classifiers. Therefore to be sure that a subset of features giving highest accuracy has been selected requires the use of many different indices.

Application of neural network for the prediction of eco-efficiency

This paper presents the application of neural networks in the design process of new technologies taking into account factors such as their influence on the environment and the economic effects of their implementation. The use of neural networks allowed eco-efficiency assessment of technologies based on highly reduced number of descriptive design parameters, which are very difficult to collect at the conceptual design stage. The great diversity of technologies involved along with the small number of available examples made difficult to construct a neural model and demanded careful data preprocessing and network structure selection.

Selection of prototype rules: context searching via clustering

Prototype-based rules are an interesting alternative to fuzzy and crisp logical rules, in many cases providing simpler, more accurate and more comprehensible description of the data. Such rules may be directly converted to fuzzy rules. A new algorithm for generation of prototype-based rules is introduced and a comparison with results obtained by neurofuzzy systems on a number of datasets provided.

Building a Model for Time Reduction of Steel Scrap Meltdown in the Electric Arc Furnace (EAF): General Strategy with a Comparison of Feature Selection Methods

Time reduction of steel scraps meltdown during the electic arc process is really a challenging problem. Typically the EAF process is stochastic without any determinism and only simple and naive rules are currently used to manage such processes. The goal of the paper is to present the way, which have been considered, to build an accurate model concerning different feature selection methods that would be helpful in predicting the end of the meltdown and maximum energy needed by the furnace.

Temperature prediction in electric arc furnace with neural network tree

This paper presents a neural network tree regression system with dynamic optimization of input variable transformations and post-training optimization. The decision tree consists of MLP neural networks, which optimize the split points and at the leaf level predict final outputs. The system is designed for regression problems of big and complex datasets. It was applied to the problem of steel temperature prediction in the electric arc furnace in order to decrease the process duration at one of the steelworks.

A model for temperature prediction of melted steel in the electric arc furnace (EAF)

A constant aspiration to optimize electric arc steelmaking process causes an increase of the use of advanced analytical methods for the process support. The goal of the paper is to present the way to predict temperature of melted steel in the electric arc furnace and consequently, to reduce the number of temperature measurements during the process. Reducing the number of temperature measurements shortens the time of the whole process and allows increasing production.

Neural network committees optimized with evolutionary methods for steel temperature control

This paper presents regression models based on an ensemble of neural networks trained on different data that negotiate the final decision using an optimization approach based on an evolutionary approach. The model is designed for big and complex datasets. First, the data is clustered in a hierarchical way and then using different level of cluster and random choice of training vectors several MLP networks are trained. At the test phase, each network predicts an output for the test vector and the final output is determined by weighing outputs of particular networks. The weights of the outputs are determined by an algorithm based on a merge of genetic programming and searching for the error minimum in some directions. The system was used for prediction the steel temperature in the electric arc furnace in order to shorten and decrease the costs of the steel production cycle.

Evolutionary optimization of regression model ensembles in steel-making process

In this paper we compare different evolutionary algorithm approaches and parameters used to optimize the output of neural network committee trained on regression problems. This is especially useful for large and complex datasets. We used the methodology presented in this paper to optimize the output of the committee to predict the temperature in the electric arc furnace in one of the steelworks.

Evolutionary optimized forest of regression trees: application in metallurgy

A forest of regression trees is generated, with each tree using a different randomly chosen subset of data. Then the forest is optimized in two ways. First each tree independently by shifting the split points to the left or to the right to compensate for the fact, that the original split points were set up as being optimal only for the given node and not for the whole tree. Then evolutionary algorithms are used to exchange particular tree subnodes between different trees in the forest. This leads to the best single tree, which although may produce not better results than the forest, but can generate comprehensive logical rules that are very important in some practical applications. The system is currently being applied in the optimization of metallurgical processes.

An algorithm of knowledge extraction from trained neural networks

The presented paper describes a method of knowledge extraction that is based on analysis of the trained ANN’s weights The method allows to determine the significance of particular inputs, to prove their synergy as well as to find some symbolic rules, that determine the direction of influence of particular inputs.