Location
260-005, Owen G. Glenn Building
Start Date
12-15-2014
Description
In the current age of big data, decision trees are one of the most commonly applied data mining methods. However, for reliable results they require up-to-date input data, which is not always given in reality. We present a two-phase approach based on probability theory for considering currency of stored data in decision trees. Our approach is efficient and thus suitable for big data applications. Moreover, it is independent of the particular decision tree classifier. Finally, it is context-specific since the decision tree structure and supplemental data are taken into account. We demonstrate the benefits of the novel approach by applying it to three datasets. The results show a substantial increase in the classification success rate as opposed to not considering currency. Thus, applying our approach prevents wrong classification and consequently wrong decisions.
Recommended Citation
Hristova, Diana, "Considering Currency in Decision Trees in the Context of Big Data" (2014). ICIS 2014 Proceedings. 1.
https://aisel.aisnet.org/icis2014/proceedings/DecisionAnalytics/1
Considering Currency in Decision Trees in the Context of Big Data
260-005, Owen G. Glenn Building
In the current age of big data, decision trees are one of the most commonly applied data mining methods. However, for reliable results they require up-to-date input data, which is not always given in reality. We present a two-phase approach based on probability theory for considering currency of stored data in decision trees. Our approach is efficient and thus suitable for big data applications. Moreover, it is independent of the particular decision tree classifier. Finally, it is context-specific since the decision tree structure and supplemental data are taken into account. We demonstrate the benefits of the novel approach by applying it to three datasets. The results show a substantial increase in the classification success rate as opposed to not considering currency. Thus, applying our approach prevents wrong classification and consequently wrong decisions.