Data stream mining is the process of extracting knowledge from continuous, rapid data records where the goal is to predict the class or value of new instances in the data stream. Real time data and events software are one of the applications most widely used by industries. The huge amount of data collected and the need to control and predict industrial business processes behavior in real time, makes that kind of software a suitable data source for applying machine learning techniques to discover processes' hidden knowledge. The goal of this work is to adapt machine learning algorithms in order to process real time data and obtain useful information for operational decision making. We will concentrate on adaptive sliding window algorithm for detecting change and keeping updated statistics from a data stream, very fast decision trees as a classification scheme, unsupervised k-means clustering for discovering data groups and cascade trained neural networks for pattern recognition. Data and knowledge visualization is made through mobile devices, desktops and tablets computers. The benefits of applying this kind of technology are: Visualize in advance events that could affect plant operation, predict and forecast equipment behavior, early fault detection, energy consumption pattern recognition, among others. Compared to traditional business intelligence software, this application has better performance for a large volume of high speed data streams, bringing real time knowledge for decision support. In order to apply the techniques exposed in this paper, a business project must be conducted to identify process knowledge needs, gaps and improvement opportunities.
展开▼
