Hoeffding Tree Algorithm is known as a method to induce decision trees from a data stream. Treatment of numeric attribute on Hoeffding Tree Algorithm has been discussed for stationary input. It has not yet investigated, however, for non-stationary input where the effect of concept drift is apparent. This paper identifies three major approaches to handle numeric values, Exhaustive Method, Gaussian Approximation, and Discretizaion Method, and through experiment shows the best suited modeling of numeric attributes for Hoeffding Tree Algorithm. This paper also experimentaly compares the performance of two known methods for concept drift detection, Hoeffding Bound Based Method and Accuracy Based Method.