Dante Conti

Mining association rules from time series to explain failures in a hot-dip galvanizing steel line

This paper presents an experience based on the use of association rules from multiple time series captured from industrial processes. The main goal is to seek useful knowledge for explaining failures in these processes. An overall method is developed to obtain association rules that represent the repeated relationships between pre-defined episodes in multiple time series, using a time window and a time lag. First, the process involves working in an iterative and interactive manner with several pre-processing and segmentation algorithms for each kind of time series in order to obtain significant events. In the next step, a search is made for sequences of events called episodes that are repeated among the various time series according to a pre-set consequent, a pre-established time window and a time lag. Extraction is then made of the association rules for those episodes that appear many times and have a high rate of hits. Finally, a case study is described regarding the application of this methodology to a historical database of 150 variables from an industrial process for galvanizing steel coils.

On the understanding of profiles by means of post-processing techniques: an application to financial assets

In last years, mining financial data has taken remarkable importance to complement classical techniques. Knowledge Discovery in Databases provides a framework to support analysis and decision-making regarding complex phenomena. Here, clustering is used to mine financial patterns from Venezuelan Stock Exchange assets (Bolsa de Valores de Caracas), and two major indexes related to that market: Dow Jones (USA) and BOVESPA (Brazil). Also, from a practical point of view, understanding clusters is crucial to support further decision-making. Only few works addressed bridging the existing gap between the raw data mining (DM) results and effective decision-making. Traffic lights panel (TLP) is proposed as a post-processing tool for this purpose. Comparison with other popular DM techniques in financial data, like association rules mining, is discussed. The information learned with the TLP improves quality of predictive modelling when the knowledge discovered in the TLP is used over a multiplicative model including interactions.

Decreasing Uncertainty When Interpreting Profiles through the Traffic Lights Panel

This paper describes Traffic Lights Panel (TLP) as a useful interpretation-oriented tool for clustering results, suitable for helping the domain experts to induce a conceptualization of the resulting profiles. Till now, the TLP is manually derived from the clustering results, but it has been well accepted by the domain experts of several real applications as a very helpful contribution to understand the classes’ meaning and improve reliable decision-making. Here, a proposal to automatically construction of TLP is presented trying to mimic the real process that the analyst performs. Two criteria based on different central trend statistics of the variables inside a class are introduced, tested with a real case study in Neurorehabilitation field and compared. Finally, uncertainty concerning TLP is analyzed; the annotated TLP (aTLP) is proposed to visualize uncertainty associated to the decisions derived from TLP, thus enhancing robustness of TLP as a supporting tool in decision-making.