Meta_LASH Tree: Bagging at Meta Level Using LASSO Regression Hoeffding Tree for Streaming Data

Building predictive model for streaming data is a major challenge as it involves high speed and huge amount of data stream which is impossible to store and process the entire data. Since the distribution of streaming data changes over time, the traditional static model is not suitable for streaming data. Nowadays most of the streaming data solutions have been centered around ensembles, which integrates predictive responses from multiple homogeneous or heterogeneous base learning algorithms. In this paper a novel, memory efficient and practically useful Ensemble Bagging at Meta level Framework denoted as Meta_LASH Tree is proposed which comprises of two phases. In the first phase a memory efficient base learner named as LASSO Regression Hoeffding Tree (LASH Tree) is constructed, which incorporates Hoeffding Tree and LASSO Regression, that produces better predictions and better insights than using both the models separately. This hybrid model is highly interpretable and can have an insights of both linear and non linear relationship of the data. In the second phase, the predictive responses from the previously constructed LASH Tree are collected using Ensemble Bagging approach, and the dominant base learner is selected by the Meta Learner. The proposed frame work is designed in such a way to reduce the memory usage and overfitting issue of the existing algorithms. It is also designed to enhance the prediction accuracy.