Visual Interpretation And Analysis Of Random Forest

Nowadays,machine learning has been successfully applied in various fields such as information retrieval,data mining,and computer speech recognition.However,due to the complexity of its functions and working mechanisms,most users in practical applications and learning regard machine learning models as black boxes,and the development of high-performance models requires a time-consuming and trial-and-error process.Therefore,the research and exploration of the interpretability of machine learning is a very important proposition.Academic researchers need more transparent and interpretable systems to better understand and analyze machine learning models.Visual analysis is an emerging technology that transforms data into informative views.It uses human' s powerful visual channels and cognitive perception capabilities to help us acquire and analyze information proactively.Therefore,we use visual analysis technology to analyze the interpretability of the model.Random forest is an integrated model composed of many independent decision trees.The overall performance of the model is better than any single decision tree.However,this also leads to poor interpretability of the model,which severely hinders the use of the model in fields that require transparency and interpretability,such as medical diagnosis.The diversity and complexity of the structure in the model is one of the biggest challenges faced by interpretation.In this article,we implemented a visual analysis system RFSeer,which contains multiple modules,multiple views,and user-friendly interactive functions.It aims to explain the structure of the random forest model in multiple dimensions,and to reduce the user's burden of understanding as much as possible.In order to prove the usability of the RFSeer system,we conducted two case studies.The main tasks completed in this thesis are as follows:(1)Design of visual analysis framework.Based on the basic theory of data visualization,this thesis introduces the design details of the visual analysis framework of the interpretable random forest in this article.This framework combines the data information in the model with the visualization means as a technical means to explore the interpretability of the random forest model.(2)The refinement of the design goals and tasks of the visual analysis system.Based on the characteristics of the random forest model and the visualization theory,the design goals and design tasks of the RFSeer system are extracted to provide guidance for subsequent design and coding work.(3)The design and realization of the view in the visual analysis system.According to the refined design goals and tasks,the modules and visual views in the design system are designed in this thesis.The design views visualize the iterative process of the model and the model structure,and finally code to implement the visual analysis system RFSeer.(4)Use the RFSeer system for interpretable case analysis.For the realized visual analysis system RFSeer,case analysis and research of two usage scenarios are carried out to prove the usability and efficiency of the system.Users can interactively understand and explore the model in multiple dimensions and from multiple angles.