Shapley Value-based Pricing For Feature Combination Of Detection Information In Cold Chain Logistics

To improve the efficiency and quality of logistics decision-making,statistical machine learning and data pricing analysis,as emerging technologies,can bring new perspectives and applications to the optimization of logistics decision-making and the process of valuable data sharing.For example,before a bank provides a loan business for a cold chain logistics company,it buys data about the cold chain logistics company to analyze the company's operating capabilities.In this process,it is necessary to price the cold chain detection information data of the transaction.However,due to the diversity,large quantity,and unclear value of this type of data,it is usually difficult to effectively determine the price of cold chain data.Therefore,according to the cold chain detection data,this thesis proposes a combination pricing model of cold chain logistics detection information based on the Shapley value to solve the problem of cold chain aquatic product shelf life prediction,the data feature contribution distribution problem in the cold chain aquatic product intelligent sorting scenario and the pricing of dynamic data.The main work and innovations are as follows:(1)Aiming at the importance of cold chain data features,the feature selection model in machine learning of cold chain detection data is studied.Based on the idea of recursive feature elimination based on cross-validation,combining feature permutation and combination,and considering prediction accuracy,design a feature selection method for cold chain detection information.The problem of feature contribution assignment based on Shapley value is studied,and the difference between the prediction result of a specific instance and the average prediction value of the data set is taken as the feature benefit of the instance.For local explanation,two random instances are drawn to simulate the“absence” of the feature.Or not,calculate the marginal contribution of the feature in a specific instance,and treat the mean of the absolute value as the global contribution of the feature in the data set.(2)Based on the cold chain logistics simulation model,the dynamic pricing problem of cold chain inspection data is studied.The transaction model of cold chain detection data is constructed,the ideal nature of long-term profit maximization is defined,the Myerson payment function and multiplication update ideas are used to derive key parameters,and a dynamic pricing model with zero regrets is proposed.The designed testing data pricing model will maximize the long-term benefits for the data provider.And realize the pricing scheme based on the multiplicative weight update algorithm,and the simulation experiment proves the feasibility of this method.(3)Apply the constructed model to the actual operation scenario of the cold chain.For the machine learning shelf-life prediction model,the proposed method can maximize the extraction of features to train the model;an analysis of a company's artificial intelligence sorting case is compared with the classic local interpretation method to prove the features based on the Shapley value The contribution allocation algorithm is effective;in addition,numerical simulation experiments have been carried out in this thesis to prove that the dynamic price update algorithm is effective.