基于改进鲸鱼优化算法的GBDT回归预测模型

针对梯度提升决策树（gradient boosting decision tree, GBDT）参数难以选择的问题, 提出一种基于改进鲸鱼优化算法（improved whale optimization algorithm, IWOA）的GBDT回归预测算法. 首先, 提出一种改进的鲸鱼优化算法, 利用混沌映射初始化种群提高种群多样性, 引入惯性权重与差分进化算法中的变异交叉策略解决迭代后期易陷入局部最优的问题； 其次, 利用IWOA对GBDT的关键参数寻优, 避免参数选择的盲目性, 提高回归预测模型的泛化能力； 最后, 建立IWOA-GBDT回归预测模型, 并利用UCI数据集对模型进行验证. 实验结果表明, 相比于决策树、 支持向量机、 Adaboost和GBDT算法, 该模型算法具有更好的拟合效果, 并有一定的实用价值.

GBDT Regression Prediction Model Based on Improved Whale Optimization Algorithm

Aiming at the problem that it was difficult to select the parameters of gradient boosting decision tree (GBDT), we proposed a GBDT regression prediction algorithm based on improved whale optimization algorithm (IWOA). Firstly, an improved whale optimization algorithm was proposed, which initialized the population by using chaotic mapping to improve the diversity of the population, and the inertial weight and the mutation crossover strategy of differential evolution algorithm were introduced to solve the problem that it was easy to fall into the local optimization in the later stage of iteration. Secondly, IWOA was used to optimize the key parameters of the GBDT to avoid the blindness of parameter selection and improve the generalization ability of the regression prediction model. Finally,  the IWOA-GBDT regression prediction model was established and verified by the UCI dataset. The experimental results show that compared with decision tree, support vector machine, Adaboost and GBDT algorithms, the proposed model algorithm has better fitting effect and certain practical value.