Line Pattern Mining Based on Density Weight Expectation Maximization and Splitting Merging Strategy

To address the issue of line pattern mining of non-linear dataset, a new regression algorithm based on density weight Expectation Maximization (EM) and splitting merging strategy is proposed. Point-direction function is first employed to establish the expression of line pattern based on finite mixture model, and grid density is introduced into EM processing as adjust weight, which can effectively reduce the possibility of fall into local optimum of regression. Then a splitting merging strategy is introduced, which ensure the proposed algorithm can overcome the connectivity limitation, and can obtain a correct result even when the number of mining is not set as the same with the real line pattern number. Experiments demonstrate that the proposed algorithm is not sensitive to the set of mining number, and is able to correctly explore the line pattern of non-linear dataset under the noise environment.