Research Of Parameter Optimization And Interpretation Method For Interval-Valued SVM

This paper mainly includes the following two aspects:(1)Optimize the parameter selection of an interval-valued SVM model;(2)Propose a data preprocessing method for LIME interpretation of interval-valued data.As one of the most widely used classification algorithms,Support Vector Machines(SVM)has good classification performance in precise-valued data sets with its rigorous mathematical logic.For interval-valued data sets,it is usually necessary to preprocess interval-valued data to transform it into precise-valued data for SVM training.The traditional method is to use the geometric center of the interval to represent the interval,and Utkin et al.proposed an interval value SVM by considering the expected risk measurement interval of the interval data,and directly introduced the interval information of the data into the training process,so that the resulting SVM The performance of the classification model is even better.However,due to the high complexity of the algorithm,it is difficult to select the parameters of the model.In view of this situation,this paper proposes a method to optimize the parameter selection of Utkin's algorithm by combining Particle Swarm Optimization algorithm and traditional method.The numerical experiments on synthesized data and 8 UCI datasets on the MATLAB platform show that the proposed method has better classification performance than the traditional method and Utkin's method.Meanwhile,the parameter optimization time of the parameter optimization method proposed in this paper in the interval-valued SVM is also much shorter than the grid search optimization method used by Utkin et al.Machine learning is widely used and is the core of many recent advances in science and technology,but most machine learning models are still black boxes,and understanding the reasons behind the predictions is important to assess the trust of the models.In general,the prediction results of the black box model can be explained by fitting a surrogate model.For the black box machine learning model with complexstructure,it is difficult to fit the satisfactory global surrogate model,so it is generally considered to use the local surrogate model to fit the single prediction of the black box model,so as to explain only the single prediction.In this paper,we use the LIME method in the local surrogate model method for interpretation.The LIME method provides an explanation for a single prediction by establishing a locally interpretable model around the prediction.However,the LIME method itself also interprets precise-valued data.Therefore,a data preprocessing method is proposed in this paper to convert the interval instances that need to be interpreted into precise instances,so that the prediction results can be directly interpreted using the LIME method.It can be seen from the numerical experiments on the Python platform that the LIME method can give an intuitive,effective and convincing explanation to the preprocessed 8 sets of UCI data.