Research On Hybrid Time Series Prediction Algorithms Based On Dynamic Ensemble Pruning

Ensemble learning demonstrates a powerful advantage in solving time series prediction problems.How to choose the appropriate base learner set from the ensemble model is the problem to be studied in ensemble pruning.Proper ensemble pruning can make the model more predictable.Ensemble pruning can effectively overcome several shortcomings of the classical ensemble learning paradigm,such as the relatively high time and space complexity.However,each predictor has its own unique ability.One predictor may not perform well on some samples,but it will perform very well on other samples.Blindly underestimating the power of specific predictors is unreasonable.Choosing the best predictor set for each query sample is exactly what dynamic ensemble pruning techniques address.This paper proposes two models suitable for solving the problem of time series prediction.The first model is a hybrid Time Series Prediction(TSP)algorithm integrating Dynamic Ensemble Pruning(DEP),Incremental Learning(IL),and Kernel Density Estimation(KDE).It dynamically selects proper predictor sets based on the kernel density distribution of all base learners' prediction values.At the same time,due to the particularity of TSP problems that samples arrive in chronological order,the idea of DEP and IL is incorporated into the algorithm.The algorithm is divided into three subprocesses: 1)Overproduction,which generates the original ensemble learning system;2)Dynamic Ensemble Pruning(DEP);3)Incremental Learning(IL).The second model is a dynamic ensemble pruning algorithm based on meta-learning.The meta-learning paradigm obtains a meta-predictor by re-learning the prediction result,and uses the meta-predictor to measure whether a base learner has the ability to predict a query sample and achieve the purpose of improving prediction performance.This paper proposes a time series prediction algorithm based on meta-learning paradigm.At the same time,three heuristic optimization algorithms are used to optimize the parameters of the algorithm,namely Genetic Algorithm(GA),Particle Swarm Optimization(PSO)algorithm and Artificial Fish Swarm Algorithm(AFSA).The algorithm can be divided into three phases: a)Ensemble Initialization;b)Meta-predictor Construction and c)Prediction.The first phase,algorithm performs initialization of the ensemble system to generate a set of base learners.In the second phase,the algorithm trains a meta-predictor to predict whether a base learner is capable of predicting a query sample.In the last phase,the trained meta-predictor is applied to the test set to obtain the final generalization error.Both algorithms have been tested on multiple time series data and compared with multiple time series prediction algorithms.Experimental results show that the two proposed algorithms have achieved good prediction performance,indicating that the proposed algorithms are suitable for time series prediction problems.