O '@ sec12-final39.pdf X [@ Research on loan prediction bas X = C [ OFile C:/Users/sunny/Documents/Data%20Mining/Research_on_loan_prediction_based_on_interpretable_machine_learning.pdf V Draw R 35F % Light rain/snow T,we R\"} Where g represents the rules mapping from samples to leaf nodes in the tree, T represents the number of leaf nodes in the tree, and W represents the score of leaf nodes. The prediction result of the ith customer ,is denoted a5 Yy (o where K is the number of iterations. 2.The formula of loss function in model training is shown in(l): Ohi= Xl )+ L) . QU=+ 11T w % G @ Where the first term of equation (1) represents the difference between the predicted value and the actual value. At the same time, to prevent overfitting due to the excessive number of leaf nodes in the tree, XGBoost sets the tree-based complexity as the regulatization term, namely formula (2) above. The objective of the solution is to find appropriate parameters that minimize the value of the loss function Obj. 3. Xgboost is an additive model, which optimizes only the submodels in the current step in each iteration, and optimizes the models generated in the previous t-1 iterations as a whole in the titeration, i.e: v 2/.(r Substituting (3) into (1), the following equation (4) can be obtained: +) Obj = 3 ILF, (x)+ [, (x). 3,1+ 2 Q1) @ 4.By Taylor expansion of equation (4), Obj function can be transformed into the following form after discarding the constant part: & 1 2 Obj = [g, fux)+ =h [2(x)]+Q(f,) = 2 5) In the top formul Q Search of5 | 1) B. SHAP As mentioned above, machine leaming algorithms are poorly interpretable, and complex integration algorithms like Xgboost are even worse, almost a black box for the user. In order to solve this problem, Lundberg et al. [11] proposed SHAP model based on Shapley value in cooperative game theory in 2017. This model can be used to explain machine learning affer the fact, and is very suitable for explaining the loan prediction model in this paper, so as o provide reference for relevant managers. Specifically, SHAP constructs an additive explanatory model that treats all features as "contributors, and its core idea is: First calculate the marginal contribution of a feature when it is added to the model, and then take the mean value of different marginal contributions of the feature in all feature sequences as the SHAPbaseline value of the feature.The ith sample in the data set is denoted as X, , the jth feature of the ith sample is denoted as x, . the predicted value of the model for this sample is denoted as |, the number of eatures of the sample is 7, and the mean value of all sample target variables is denoted as y, . . then the caleulation formula of SHAP value is: B Ve 4 L) 5] e f5,) (6) where /(x,) is the SHAP value of , which is the contribution value of the jth feature of the ith sample to the final prediction result. This value can reflect the change of the prediction result when the characteristic is taken as the condition, that is, the impact on the prediction result. Therefore, we can use SHAP theory to explain and analyze each feature of the lending user and calculate the impact of each feature on the final prediiction. C. General model framework %' 'L";: . \\.._ ousy B 8 == :::::;xv T k' Figurel Framework 1L The experiment A. The hardware and software environment of the experiment The hardware configuration of this experiment is: i5- 11320H 3.20GHz processor and Iris(R) Xe Graphics Graphics Graphics_graphics card, the experiment is carried out in Python3.7.4 and jupyter notebook environment. B. The evaluation indlex of the experiment The evaluation indexes of this experiment are Accuracy score, Precision score, Recall score and FI score. Among them, S & a ) PR Edit with Acrobat 4:20 PM 12/2/2024 a O [@ sec12-final39.pdf X [@ Research on loan prediction bas X = C [ OFile C:/Users/sunny/Documents/Data%20Mining/Research_on_loan_prediction_based_on_interpretable_machine_learning.pdf \"E" V Draw A 35F Light rain/snow