The Prediction Of Time Spend On Automobile Test System Using Regression Analysis

With the rapid development of the science and technology,the innovation of automobile industry plays an important role.In order to ensure the safety and reliability of car configuration before going on the road,engineers in automobile companies generally need to develop a powerful test system.However,without the support of algorithm,it is complex and time-consuming to optimize the speed of the test system for such a variety of function combinations.Therefore,an effective algorithm is particularly important for prediction of the time required for a car to pass the test system.This thesis devotes to a descriptive statistical analysis of the anonymous vehicle test data of Benz company,and builds a model by XGBoost.The model uses crossvalidation method,and evaluates the model with the coefficient of determination(R~2)and mean square error(MSE)as the main indicators.For categorical variables,label encoding and one-hot encoding are used to analyze their impact on the performance of the model.Moreover,the encoding method is further improved by target encoding.After determining the optimal encoding method,the feature selection is carried out by null importance,which further improves the accuracy of the model.