A Spark-based Ant Lion Algorithm for Parameters Optimization of Random Forest in Credit Classification

摘要

According to the fact that in large scale credit classification there is no suitable method for determining the parameter combination when random forest algorithm is multi-calss classification to get the optimal classification result. Based on Spark, this paper proposed a parallel Ant Lion Optimizer(ALO) algorithm to seek optimal parameter combination. Compared with the traditional stand-alone ALO algorithm, the Spark-based parallel strategy initially divided the whole ant population into several sub-populations, then assigned one sub-population to one partition in the RDD, and specified independent evolution in one partition. Finally it utilized the migration operator to exchange information between sub-populations. In the experiment, the lending club loan data was used to classify the borrower's loan classification. It concluded that the Spark-based parallel Ant Lion Optimizer (SALO) algorithm, compared with traditional Ant Lion Optimizer algorithm, can efficiently find the better parameter combinations to improve the classification accuracy in the random forest parameter tuning process. In addition, under the big data distributed Spark platform, the SALO whose speed is so fast and acceleration effect is so obvious will be used as the next-generation parameter optimizer of the cloud computing platform.