An Interactive Predictive Data Mining System for Informed Decision
There exists a need to utilize the predictive data mining models for querying to obtain the predicted outcome based on user provided inputs in its real use. This demo illustrates a real-world situation in which the trained predictive data mining system is
- PDF / 133,571 Bytes
- 4 Pages / 430 x 660 pts Page_size
- 15 Downloads / 237 Views
Abstract. There exists a need to utilize the predictive data mining models for querying to obtain the predicted outcome based on user provided inputs in its real use. This demo illustrates a real-world situation in which the trained predictive data mining system is being deployed and now users can interact with the model for informed decision. Keywords: Data Mining, predictive model, interactive model.
1 Introduction The training of a predictive data mining model and then understanding rules and patterns inferred by the mining model should not be the end of the prediction task. The data mining model should allow the user to query the system for future cases. The created data mining model needs to be deployed in practice as a user-driven prediction system so that the data mining system can be used for querying to obtain the predicted outcome based on user provided inputs. In some real-world applications, the training set contains relatively complete attribute information while the unseen cases (user queries) do contain many missing attribute values. Consider a predictive data mining model that is built to predict the “Service Life” of building components based on the input attributes such as “Location”, “Component”, “Material”, “Salt Deposition”, and “Mass Loss”. Suppose a builder (a typical user of the predictive model/tool/system) wants to know the service life of a “Gutter” with “Galvanized Steel” at a particular location. The user does not explicitly know the “Salt Deposition” and “Mass Loss” in that location. The user query will include two missing values. In such a case, the predicted service life by the predictive data mining tool will not be as accurate as tested in the evaluation phase of the predictive model, especially when the missing attributes play key roles in predicting the outcome. On the other hand, if the “Salt Deposition” and “Mass Loss” features are excluded from the model building, the performance of the model may not be acceptable. Hence, a major problem that needs to be solved is how to select the appropriate attributes to build the model for real-world situations where the users can not provide all the inputs for querying to the system. In other words, how to deal with the missing attribute values in user queries (unseen cases). We developed an interactive data mining model for predicting the J.R. Haritsa, R. Kotagiri, and V. Pudi (Eds.): DASFAA 2008, LNCS 4947, pp. 694–697, 2008. © Springer-Verlag Berlin Heidelberg 2008
An Interactive Predictive Data Mining System for Informed Decision
695
service life of metallic components in buildings which allows the user to input the queries based on their limited knowledge, while maintaining the accuracy of the predicted outcome.
2 An Interactive Predictive Data Mining System The proposed interactive predictive data mining system consists of nine phases structured as sequences of predefined steps (as shown in Figure 1). The system includes the standard data pre-processing, data analysis and result post-processing phases. Additionally, it include
Data Loading...
