Abstract:
|
In the new spline regression metho, the knots of splines are selected by using an univariate regression processing. First, a portion of the predictor points is picked up for the initial regression processing as seed group. The second group of regression as a comparison group is that extra points are added into the first group. Then, the regression results between these two groups are compared by using some criteria. If the extra points do not cause the decrease of regression criteria exceeding some limit, the comparison group becomes the seed group and then repeats the previous processing. The right point in the last seed group is defined as the knot if the processing fails. The processing for the remain points of predictor is to repeat until all knots of the predictor are found. In the regression processing, you can use either linear regression function or binomial function. A serious comparison studies between the new method with other data mining methods is also conducted on a binary outcome prediction problem by using Hosmer-Lemeshow good fitness test. The result of the comparison shows that the new method has better result in prediction, efficiency, and stability.
|