Knowledge Extraction Process Research Articles

A Data Mining Approach For Medium-term Demand Forecasting

The Brazilian government, in the past, monopolized the generation, transmission and distribution of electric energy. Following world-wide tendencies, nowadays the government assumes the regulation function in a competitive and horizontal market. The variable load, fundamental in the planning of the electrical and energy operation, in the studies of magnifying and reinforcement of the basic network, assumes strategic importance in the commercial area, improving the data storage and knowledge extraction process using computational techniques. In the present work, data mining techniques are used to produce monthly load forecasts in intervals of high, medium and low consumption, according to the needs of electrical energy distribution companies. The results of neural network models, when compared to statistical results, show improved performance, presenting an Average Relative Error about 0.5% lower.

Velocity predictions in compound channels with vegetated floodplains using genetic programming

Data collection and storage methods have improved vastly over recent years, however the processes of information and knowledge extraction from data have not mirrored this. The application of computer supported scientific knowledge discovery processes to carefully collected observations aims to improve the understanding of the processes that generated or produced these data. In this paper, these new techniques have been applied to the complex and poorly understood phenomena of flow through idealised vegetation. The ability to predict, with improved accuracy, velocities within wetlands and other vegetated areas would be advantageous as these regions are increasingly being recognised for their natural flood alleviation properties. In this study, laboratory data collected in a flume with steady flows over a deep channel with relatively shallow vegetated floodplains were used to induce the formulation of expressions using a data driven discovery technique, namely genetic programming (GP). The objective of the study was not only to gain an understanding of the effect of vegetation on velocity distributions across a channel but moreover to demonstrate an alternative discovery process. The performance of the genetic program is reported for three variations of the GP. The reported results of the experiments were found to be encouraging and further work is detailed.

Decomposition in data mining: an industrial case study

Data mining offers tools for discovery of relationships, patterns, and knowledge in large databases. The knowledge extraction process is computationally complex and therefore a subset of all data Is normally considered for mining. In this paper, numerous methods for decomposition of data sets are discussed. Decomposition enhances the quality of knowledge extracted from large databases by simplification of the data mining task. The ideas presented are illustrated with examples and an industrial case study. In the case study reported in this paper, a data mining approach is applied to extract knowledge from a data set. The extracted knowledge is used for the prediction and prevention of manufacturing faults in wafers.

Evolving Fuzzy Rule-Based Models

An approach of evolving fuzzy rule-based models by genetic algorithms (GA) is proposed in the paper. Both structure of the model (fuzzy rules) and parameters of the fuzzy membership functions of the linguistic variables are generated automatically. The main feature of the proposed approach is the new encoding mechanism of the chromosome that is more efficient than encoding used in previous evolutionary learning methods. The representation does not need all the rules to be present because the GA selects a small subset of the used rules only. This fact leads to minimizing the computational load using significantly smaller chromosome and real-coded GA, making possible simultaneous parameter and structural identification. Evolving fuzzy rule-based models need only inputs and outputs to be known, but unlike the other typical black-box models'(neura1 networks, polynomial models etc.) their transparency is very high due to the design of linguistic rules during the process of knowledge extraction and aggregation. Two practical building services engineering problems are considered in order to illustrate the applicability of the approach.

Knowledge discovery in data warehouses

As the size of data warehouses increase to several hundreds of gigabytes or terabytes, the need for methods and tools that will automate the process of knowledge extraction, or guide the user to subsets of the dataset that are of particular interest, is becoming prominent. In this survey paper we explore the problem of identifying and extracting interesting knowledge from large collections of data residing in data warehouses, by using data mining techniques. Such techniques have the ability to identify patterns and build succinct models to describe the data. These models can also be used to achieve summarization and approximation. We review the associated work in the OLAP, data mining, and approximate query answering literature. We discuss the need for the traditional data mining techniques to adapt, and accommodate the specific characteristics of OLAP systems. We also examine the notion of interestingness of data, as a tool to guide the analysis process. We describe methods that have been proposed in the literature for determining what is interesting to the user and what is not, and how these approaches can be incorporated in the data mining algorithms.

INSS: A hybrid system for constructive machine learning

In this paper we present the INSS system, a new hybrid approach based upon the principles of KBANN networks. It represents an important improvement in comparison with its predecessor because the learning and the knowledge extraction process are faster and are accomplished in an incremental way. INSS offers a new approach applicable to constructive machine learning with high-performance tools, even in the presence of incomplete or erroneous data.

Extracting finite-state representations from recurrent neural networks trained on chaotic symbolic sequences

While much work has been done in neural-based modeling of real-valued chaotic time series, little effort has been devoted to address similar problems in the symbolic domain. We investigate the knowledge induction process associated with training recurrent neural networks (RNN's) on single long chaotic symbolic sequences. Even though training RNN's to predict the next symbol leaves the standard performance measures such as the mean square error on the network output virtually unchanged, the networks nevertheless do extract a lot of knowledge. We monitor the knowledge extraction process by considering the networks stochastic sources and letting them generate sequences which are then confronted with the training sequence via information theoretic entropy and cross-entropy measures. We also study the possibility of reformulating the knowledge gained by RNN's in a compact and easy-to-analyze form of finite-state stochastic machines. The experiments are performed on two sequences with different "complexities" measured by the size and state transition structure of the induced Crutchfield's epsilon-machines. We find that, with respect to the original RNN's, the extracted machines can achieve comparable or even better entropy and cross-entropy performance. Moreover, RNN's reflect the training sequence complexity in their dynamical state representations that can in turn be reformulated using finite-state means. Our findings are confirmed by a much more detailed analysis of model generated sequences through the statistical mechanical metaphor of entropy spectra. We also introduce a visual representation of allowed block structure in the studied sequences that, besides having nice theoretical properties, allows on the topological level for an illustrative insight into both RNN training and finite-state stochastic machine extraction processes.

Application of machine learning in water distribution networks

This paper describes an ongoing work on the application of machine learning techniques in the domain of water distribution networks. This research is performed in the framework of the European project WATERNET, whose aim is to develop a system to control and manage water distribution networks. WATERNET is composed of a supervision system, a distributed information management subsystem, an optimization subsystem, a water quality monitoring subsystem, and a simulation subsystem. In addition to these components, a machine learning subsystem is included to extract knowledge from historical data and improve the performance of the water management system. This paper is focused on the approach and methodology followed for the development of the machine learning subsystem. The basic raw material for this work are historical data from a Portuguese water distribution company that has 45 water stations and some of them with six years of data collected every five minutes. The paper also shows the first results obtained, discusses difficulties found in the performed experiments and introduces an architecture based on qualitative models/causal relationships to ease the process of knowledge extraction from the historical data and the assessment of the extracted knowledge.

Selection of Lawn and Garden Tractors Using an Expert System

NO known guidelines have existed for consumer selection of lawn and garden tractors. Certain rules for selection were developed from a knowledge extraction process from experts (top salespersons) for two major brands of outdoor products. Their knowledge was used in constructing expert systems to assist consumer selection. Each brand required its own unique programming logic in order to accomplish proper selection. The accuracy of the expert systems was verified by comparing the model output to salesmens' selections.

Expert System for Interpreting Citrus Harvest Simulation

A decision support system was developed to aid citrus harvest managers in properly interpreting simulation results and developing good management strategies. A study of Florida citrus harvest managers revealed several areas of misinterpretation in the output from a harvest simulation model. These were attributed to lack of proper understanding of the concepts presented as well as lack of necessary statistical training. A knowledge extraction process involving experts in simulation analysis was used to develop rules and logic for interpreting simulation results. An expert system shell was used to capture the rules and logic employed by the expert. The resulting expert system was then linked to the simulation model to produce a decision support system.

Learning design rules from decisions and performances

This paper examines an approach to the extraction of implicit knowledge in rule form about the relationships between design decisions and their performance consequences. The effects of an imposed structure on a performance space are observed in relation to matching points in a decision space. A mapping between the two spaces embodies the knowledge that is discovered. The performance space is structured by Pareto optimization and the knowledge extraction process is illustrated by two examples from building design. The use of the methodology for learning about decision/performance relationships in extant designs is proposed.