Research On Instance Selection Of Structured Data Based On Reinforcement Learning

With the increase of data acquisition channels,dimensions and quantity,the data set that people collect directly may contain many instances that affect the quality of data set,such as noise,untruth,redundancy,etc.The existence of these examples will have a negative impact on the subsequent machine learning model training,and even reduce the performance of the decision model.Therefore,if we can choose a better data subset through a reasonable example selection,it is possible to avoid this problem and make the data play its due value.Based on this scenario,this paper proposes a structured data instance selection method based on reinforcement learning.From the perspective of improving the performance of the model,this method transforms the problem of case selection into the design of reinforcement learning environment,and realizes the complete process of case selection with the help of the relevant algorithms in reinforcement learning.In the aspect of concrete implementation,this method designs a complete reward feedback mechanism by linking the modeling performance of the screening set with the reward signal of reinforcement learning,so that the reward signal can reflect the effect of building machine learning model based on this screening set.At the same time,it can be used in machine learning modeling,data mining and other application scenarios as part of data preprocessing.In order to verify the effectiveness of this method,this paper uses the artificial data set and the public data set to analyze the interpretation and verify the performance of this method.The experimental results show that this method can improve the performance of the model by eliminating the useless instances,and shows good stability.Compared with the existing case selection methods,this method has some advantages in improving the performance of filtering set modeling,and has a high value of use.