Hélder Quintela

Corporate bankruptcy prediction using data mining techniques

The interest in the prediction of corporate bankruptcy is increasing due to the implications associated with this phenomenon (e.g. economic, and social) for investors, creditors, competitors, government, although this is a classical problem in the financial literature. Two kinds of models are generally adopted for bankruptcy prediction: (i) accounting ratios based models and (ii) market based models. In the former, classical statistical techniques such as discriminant analysis or logistic regression models have been used, while in the latter the Moody’s KMV model was adopted. This paper follows the first approach (i), and it is based on the analysis of the evolution of several financial indicators during a three-year period. A framework was developed, encompassing a total of 16 models. These differ in the data mining algorithm (e.g. Artificial Neural Networks or Decision Trees), the data used (all three years or just the last one) and the input attributes adopted (e.g. all accounting ratios or just the most significant ones). The experiments were conducted using the new Business Intelligence Development Studio of the Microsoft SQL Server. Very good results were achieved, with performances between 86% and 99% for all 16 models.

Ecological mining: a case study on dam water quality

The automatic assessment of barrage water quality is very restricted due to the distances, the number of biochemical parameters to be considered and the financial resources spent to obtain their values. To this scenario should be added the latency times between the sampling moment and the outcome of the laboratory analyses. Although the idea of considering sensors for remote acquisition of data is not new, there are some constraints to be addressed, like the existence of sensors to measure the pertinent parameters and their efficiency, the costs involved and the possibility of remote sensing. The application of this alternative is highly dependent on the relevance of the candidate parameters. At this point, the Data Mining (DM) approach assumes an important role, in the sense that it can reveal the relative importance of the parameters, as well the prediction models to determine the water quality and finally the associated accuracies. This paper introduces a decision framework to support the selection of biochemical parameters to be considered in remote sensing of water contained in barrages. The framework enables the comparison of the efficiency of two kinds of models, using decision trees. The first one uses all the water quality indicators, including the time and cost consuming variables, while the second model is based only on remotely real-time acquired parameters. When comparing both strategies under several criteria (e.g., cost, time and confidence), the latter method was showed to be the best alternative.

Real-time intelligent decision support system for bridges structures behavior prediction

There is an increasing need of deploying automatic real-time decision support systems for civil engineering structures like bridges, making use of prediction models based in Artificial Intelligence techniques (e.g., Artificial Neural Networks) to support the monitoring and prediction activities. Past experiments with Data Mining (DM) techniques and tools opened room for the development of such a real-time Decision Support Systems. However, it is necessary to test this approach in a real environment, using real-time sensors monitoring. This study presents the development of prediction models for structures behavior and a novel architecture for operating in a real-time system.

An intelligent decision support system for bridge safety assessment based on Data Mining models

Serie : WIT transactions on information and communication technologies, vol. 40, ISSN 1746-4463