LASH Tree: LASSO Regression Hoeffding for Streaming Data

Streaming data is a challenging research area for the last two decades which comes in high volume and rapid speed and cannot be stored using existing memory. Dealing with model adaptability with evolving data over time and memory usage arethe major challenges in streaming data predictive models. Recently there is a rising attention in developing Regression Tree models due to it’s high interpretability and accuracy. Additionally, the linear function at the leaf node evaluates the target variable more accurately by analysing the correlation between predictor variables and target variable. The proposed LASSO Regression Hoeffding Tree (LASH Tree) is a Regression Tree model which incorporates LASSO Regression with Hoeffding Tree that produces better predictions and better insights. In this paper, an exhaustive empirical testing of the proposed methodology is performed and compared with other standard model like CART, Hoeffding based Linear Regression Model (ORTO) using solar energy data set. The obtained results show that the proposed LASH Tree significantly outperforms the existing approaches and it is proved that there is boosting of accuracy and usedless memory usage when compared with other algorithms.