Search results for "Data mining"
showing 10 items of 907 documents
Dynamic integration of classifiers in the space of principal components
2003
Recent research has shown the integration of multiple classifiers to be one of the most important directions in machine learning and data mining. It was shown that, for an ensemble to be successful, it should consist of accurate and diverse base classifiers. However, it is also important that the integration procedure in the ensemble should properly utilize the ensemble diversity. In this paper, we present an algorithm for the dynamic integration of classifiers in the space of extracted features (FEDIC). It is based on the technique of dynamic integration, in which local accuracy estimates are calculated for each base classifier of an ensemble, in the neighborhood of a new instance to be pr…
Multidimensional Model Design using Data Mining: A Rapid Prototyping Methodology
2017
[Departement_IRSTEA]Ecotechnologies [TR1_IRSTEA]MOTIVE; International audience; Designing and building a Data Warehouse (DW), and associated OLAP cubes, are long processes, during which decision-maker requirements play an important role. But decision-makers are not OLAP experts and can find it difficult to deal with the concepts behind DW and OLAP. To support DW design in this context, we propose: (i) a new rapid prototyping methodology, integrating two different DM algorithms, to define dimension hierarchies according to decision-maker knowledge; (ii) a complete UML Profile, to define a DW schema that integrates both the DM algorithms; (iii) a mapping process to transform multidimensional …
A Methodology and Tool for Rapid Prototyping of Data Warehouses Using Data Mining: Application to Birds Biodiversity
2014
Data Warehouses (DWs) are large repositories of data aimed at supporting the decision-making process by enabling flexible and interactive analyses via OLAP systems. Rapid prototyping of DWs is necessary when OLAP applications are complex. Some work about the integration of Data Mining and OLAP systems has been done to enhance OLAP operators with mined indicators, and/or to define the DW schema. However, to best of our knowledge, prototyping methods for DWs do not support this kind of integration. Then, in this paper we present a new prototyping methodology for DWs, extending [3], where DM methods are used to define the DW schema. We validate our approach on a real data set concerning bird b…
Convolutional Matrix Factorization for Recommendation Explanation
2018
In this paper, we introduce a novel recommendation model, which harnesses a convolutional neural network to mine meaningful information from customer reviews, and integrates it with matrix factorization algorithm seamlessly. It is a valid method to improve the transparency of CF algorithms.
Virtual lock-and-key approach: The in silico revival of Fischer model by means of molecular descriptors
2010
Abstract In the last years the application of computational methodologies in the medicinal chemistry fields has found an amazing development. All the efforts were focused on the searching of new leads featuring a close affinity on a specific biological target. Thus, different molecular modeling approaches in simulation of molecular behavior for a specific biological target were employed. In spite of the increasing reliability of computational methodologies, not always the designed lead, once synthesized and screened, are suitable for the chosen biological target. To give another chance to these compounds, this work tries to resume the old concept of Fischer lock-and-key model. The same can …
Comparing Recurrent Neural Networks using Principal Component Analysis for Electrical Load Predictions
2021
Electrical demand forecasting is essential for power generation capacity planning and integrating environment-friendly energy sources. In addition, load predictions will help in developing demand-side management in coordination with renewable power generation. Meteorological conditions influence urban area load pattern; therefore, it is vital to include weather parameters for load predictions. Machine Learning algorithms can effectively be used for electrical load predictions considering impact of external parameters. This paper explores and compares the basic Recurrent Neural Networks (RNN); Simple Recurrent Neural Networks (Vanilla RNN), Gated Recurrent Units (GRU), and Long Short-Term Me…
The lift computation for an oscillating flat plate in incompressible potential flow
1994
The initial aim of this work was the estimation of the lift acting on a flat plate performing small oscillations in a plane uniform stream by means of a simplified model based on one or at the most two lumped vortices, and the assessment of its results by comparison to those that were exact. The model was found to work well up to a reduced frequency of about 1 or 2, above which the results diverged from those that were correct. In order to improve the model, its behaviour at very high frequencies was then investigated, discovering: (i) that if the number of lumped vortices is greater than one the possibility to impose all boundary conditions is subject to certain geometrical constraints; (i…
Cost-effective Multiresolution schemes for Shock Computations
2009
Harten's Multiresolution framework has provided a fruitful environment for the development of adaptive codes for hyperbolic PDEs. The so-called cost-effective alternative [4,8,21] seeks to achieve savings in the computational cost of the underlying numerical technique, but not in the overall memory requirements of the code. Since the data structure of the basic algorithm does not need to be modified, it provides a set of tools that can be easily implemented into existing codes and that can be very useful in order to speed up the numerical simulations involved in the testing process that is associated to the development of new numerical schemes. In this paper we present two different applica…
An Empirical Evaluation of Common Vector Based Classification Methods and Some Extensions
2008
An empirical evaluation of linear and kernel common vector based approaches has been considered in this work. Both versions are extended by considering directions (attributes) that carry out very little information as if they were null. Experiments on different kinds of data confirm that using this as a regularization parameter leads to usually better (and never worse) results than the basic algorithms.
Forecast of Study Success in the STEM Disciplines Based Solely on Academic Records
2020
We present an approach to the forecast of the study success in selected STEM disciplines (computer science, mathematics, physics, and meteorology), solely based on the academic record of a student so far, without access to demographic or socioeconomic data. The purpose of the analysis is to improve student counseling, which may be essential for finishing a study program in one of the above mentioned fields. Technically, we show the successful use of propositionalization on relational data from educational data mining, based on standard aggregates and basic LSTM-trained aggregates.