Extended Abstract (80K)3.7
Ensemble classifier for winter storm precipitation in polarimetric radar data
Emmanuel Goossaert, University of Oklahoma, Norman, OK; and R. Alam
The precipitation classification problem for this contest consists of classifying data instances into three class types: frozen, liquid and none. Based on our experiments on the data set using available open source classifiers, we observed that different algorithms have different precisions for each class type. Thus, we decided to develop two classifiers adapted to the data set: Neural Network and Decision Tree. Our implemented solutions had similar performance issues as the open source classifiers. Since individual classifiers are unable to provide high accuracy classification for all of the class types, we chose to combine the best classifiers using a multi-class linear ensemble technique. This technique allows the different classifiers to vote based on their accuracy and precision for each class type. The obtained solution performs better compared to the individual classifiers that compose it. These classifiers are: Decision Tree C4.5 and J48, Neural Network, Naive Bayes, Naive Bayes Tree, Bayes Network, Random Forest, OneR and SMO.
Session 3, Forecasting Contest
Tuesday, 13 January 2009, 3:30 PM-5:30 PM, Room 125A
Previous paper Next paper
Recorded presentation