Risk intelligent monitoring method for guarantee-deposit-free online leasing customers

Abstract

The invention provides a guarantee-deposit-free online lease customer risk intelligent monitoring method, and relates to the technical field of guarantee-deposit-free online lease, the guarantee-deposit-free online lease customer risk intelligent monitoring method comprises the following steps: collecting a historical lease user data set; constructing a decision tree model by using the data set of the historical leasing users; collecting personal data and external environment data of a new leasing user during a leasing period, and calculating an overdue renting risk probability of the new leasing user through the constructed decision tree model; and in combination with the overdue renting risk probability of the new renting user, the new renting user is divided into corresponding risk levels, and corresponding precautionary measures and collection strategies are formulated according to the risk levels of the new renting user. Therefore, the overdue renting rate of new renting users can be reduced, the potential operation risk is reduced, the renting platform can better control the credit risk, the problems of financial crisis, business stagnation and the like possibly caused by a large amount of overdue time are reduced, the brand reputation of the renting platform is improved, and the healthy development of the renting market is promoted.

Description

Technical Field

[0001] This application relates to the technical field of online rental without deposit, and particularly to an intelligent monitoring method for customer risks in online rental without deposit. Background Art

[0002] Online rental of mobile phones without deposit means that users can rent mobile phones through an online platform without paying the deposit required by the traditional rental mode. After passing the credit assessment of the rental platform, users can select and rent the mobile phones they need. During the rental period, users need to pay the rent as agreed, and after the rental period expires, they can choose to renew the lease, buy out, or return the mobile phone. Monitoring during the rental period refers to the whole-process management from the entry into force of the lease contract, the start of the rental behavior until the end of the lease contract and the return of the leased item. During the customer's rental period, it is necessary to monitor the possible existing risks and give early warnings to users.

[0003] For example, the invention patent with the publication number CN117726416A discloses a method for online rental of mobile phones. This method collects users' online behavior data, and the behavior data includes users' browsing history, shopping habits, and application preferences; analyzes and processes the collected users' behavior data to predict users' mobile phone usage needs; recommends mobile phone models suitable for their needs according to the predicted mobile phone usage needs; determines the rental price of the recommended mobile phones according to the mobile phone models, market demand, users' credit ratings, and historical rental records. During the process of users using the rented mobile phones, by collecting the status data of the mobile phones, potential risks are identified and corresponding early warning notifications are sent to users. By real-time monitoring of the key status data of the rented mobile phones and combining effective risk identification and timely early warning notifications, an enhanced security and protection mechanism is provided for users.

[0004] However, when the above method for online rental of mobile phones monitors the risks during the rental period, it only monitors the motion state data, temperature data, and humidity data of the mobile phones, as well as the light intensity data and sound intensity data of the environment, and cannot monitor the behavior of users. This may lead to an increase in the overdue rent rate of users, and further result in the rental platform being unable to recover the rent on time, the capital turnover being blocked, affecting the normal turnover and reinvestment of funds. If a large number of users are overdue, it may also cause the rental platform to be affected by a shortage of funds, which will affect the further expansion of the business or the increase in operating costs, and an excessively high overdue rent rate of users may also have an adverse impact on the market and damage the brand reputation of the rental platform. Summary of the Invention

[0005] The purpose of this application is to provide an intelligent monitoring method for customer risks in online rental without deposit, which is used to solve the problem that the method for online rental of mobile phones in related technologies cannot monitor the behavior of users, thus possibly leading to an increase in the overdue rent rate of users.

[0006] The intelligent risk monitoring method for online rental customers without deposit provided by this application adopts the following technical solutions:

[0007] An intelligent risk monitoring method for online rental customers without deposit, comprising:

[0008] Collecting a historical rental user data set, which includes personal data, external environment data, and corresponding overdue rent payment situations during the user's sample rental period;

[0009] Using the data set of historical rental users to construct a decision tree model;

[0010] Collecting personal data and external environment data during the rental period of new rental users, and calculating the overdue rent risk probability of new rental users through the constructed decision tree model;

[0011] Combining the overdue rent risk probability of new rental users, classifying new rental users into corresponding risk levels, and formulating corresponding preventive measures and collection strategies according to the risk levels of new rental users.

[0012] Optionally, the steps of using the data set of historical rental users to construct a decision tree model, collecting personal data and external environment data during the rental period of new rental users, and calculating the overdue rent risk probability of new rental users through the constructed decision tree model include:

[0013] Calculating the information entropy H(D) of the historical rental user data set D through the following formula:

[0014]

[0015] where c is the number of categories, and p i is the proportion of the i-th type of sample in the data set D;

[0016] Calculating the conditional entropy H(D|A) of a certain feature A in the historical rental user data set D through the following formula:

[0017]

[0018] where n is the number of values of the feature A, D j represents the subset where the value of the feature A is j, |D j | and D| respectively represent the number of samples in the subset D j and the data set D, and H(D j ) is the information entropy of the subset D j ;

[0019] Calculating the information gain Gain(D, A) of each feature in the historical rental user data set D through the following formula:

[0020] Gain(D,A) = H(D) - H(DA),

[0021] where H(D) is the information entropy of the historical rental user dataset D, and H(DA) is the conditional entropy of a certain feature A in the historical rental user dataset D;

[0022] Select the feature with the largest information gain Gain(D,A) as the splitting feature of the current decision tree node, and build the decision tree accordingly;

[0023] For new rental users, starting from the root node of the decision tree, follow the corresponding branches downward according to their corresponding feature values, and finally reach a certain leaf node. Calculate the overdue rent risk probability of the new rental user at the leaf node through the following formula

[0024]

[0025] where m is the number of users with overdue rent in the sample set included in the reached leaf node, and N is the total number of rental user samples at the leaf node.

[0026] Optionally, the personal data includes age, occupation, income situation, marital status, credit assessment data, rent payment behavior data, education level, residential stability, family financial situation, consumption behavior data, social network data, financial market data, and loan contract data; the external environment data includes macroeconomic conditions, industry trends, and regional difference situations.

[0027] Optionally, after collecting the personal data and external environment data during the rental period of historical rental users and new rental users, clean the collected personal data and external environment data respectively, remove data records with duplicates, errors, and excessive missing values, and discretize continuous data according to the actual situation.

[0028] Optionally, when using the dataset of historical rental users to build a decision tree model, randomly sample with replacement from the historical rental user dataset D to obtain K sub-training datasets D 1 , D 2 ,... D K , where N is the number of samples in the historical rental user dataset D;

[0029] In each sub-training dataset D i , randomly select a feature subset F of size m from the entire feature set F i , where or m = log 2 |F|;

[0030] Use each sub-training dataset D iand the corresponding feature subset F i Train a decision tree T respectively i , and obtain a random forest model consisting of K decision trees {T 1 , T 2 ,..., T k};

[0031] When calculating the overdue rent risk probability of a new rental user, input it into each decision tree T i for prediction, and obtain K prediction results and use the average method to obtain the overdue rent risk probability of the new rental user

[0032] Optionally, set a threshold for the proportion of the number of overdue rent samples in the total number of samples. When the proportion of the number of overdue rent samples in the historical rental user dataset collected is lower than the set threshold, for each overdue rent sample in the historical rental user dataset, find the k samples x i nearest to the overdue rent sample in its feature space i1 , x i2 ,..., x ik , randomly select one of the nearest neighbor samples x ij , and generate a new overdue rent sample x new by linear interpolation, and the formula for generating the new overdue rent sample x new is as follows:

[0033] x new = x i + λ(x ij - x i ),

[0034] where λ is a random number between 0 and 1.

[0035] Optionally, when using the historical rental user dataset to construct a decision tree model, divide the historical rental user dataset into two feature subsets F 1 and F 2 , use the feature subset F 1 and the corresponding customer samples to train the decision tree model, use the feature subset F 2 and the corresponding customer samples to train the logistic regression model, respectively predict the overdue rent risk probability of the new rental user through the decision tree model and the logistic regression model, and fuse the prediction results of the two models to obtain the fused overdue rent risk probability.

[0036] Optionally, the hypothesis function of the logistic regression model is where h θ (x) is the output of the hypothesis function, where x is based on the feature subset F2 The customer feature vector, θ is the parameter vector of the logistic regression model, e is the base of the natural logarithm, -θ T x is the inner product of the parameter vector θ and the feature vector x;

[0037] The parameter θ is estimated by the maximum likelihood estimation method to maximize the likelihood function of the logistic regression model on the training set. After training, for the new lease user sample x new , the prediction is made through the decision tree model and the logistic regression model respectively to obtain the overdue rent risk probability of the new lease user and

[0038] Optionally, the prediction results of the two models are fused to obtain the fused overdue rent risk probability, including: fusing the overdue rent risk probability of the new lease user according to the following formula and

[0039] where, is the fused overdue rent risk probability, w 1 is the weight of the decision tree model, w 2 is the weight of the logistic regression model, and w 1 +w 2 = 1, is the overdue rent risk probability of the new lease user predicted by the decision tree model, is the overdue rent risk probability of the new lease user predicted by the logistic regression model.

[0040] Optionally, during the lease period of the new lease user, the status data of the mobile phone is collected. The status data includes the power-on duration P usage of the mobile phone, the abnormal location situation P location and the mobile phone repair situation P maintenance . The overdue rent probability P phone of the mobile phone status of the new lease user is calculated through the following formula

[0041] P phone = β × P usage + γ × P location + δ × P maintenance ,

[0042] where, β, γ, δ are weight coefficients, and β + γ + δ = 1;

[0043] The overdue rent risk probability of the new lease user is calculated through the constructed decision tree model Combined with the overdue rent risk probability and the overdue rent probability P phone, and calculate the comprehensive overdue rent probability P through the following formula:

[0044] Among them, α is the weight coefficient;

[0045] Divide the new lease users into corresponding risk levels according to the comprehensive overdue rent probability P.

[0046] To sum up, this application at least includes the following beneficial technical effects: The intelligent risk monitoring method for deposit-free online lease customers of this application collects the historical lease user dataset, constructs a decision tree model using the dataset of historical lease users, then collects the personal data and external environment data of new lease users during the lease period of the machine, calculates the overdue rent risk probability of new lease users through the constructed decision tree model, and then combines the overdue rent risk probability of new lease users to divide the new lease users into corresponding risk levels, and formulates corresponding preventive measures and collection strategies according to the risk levels of new lease users, so as to reduce the overdue rent rate of new lease users, reduce potential operation risks, help the lease platform better control credit risks, make the lease business operate more steadily, reduce problems such as financial crises and business stagnation that may be caused by a large number of overdue payments, improve the brand reputation of the lease platform, and promote the healthy development of the lease market. Description of the Drawings

[0047] Figure 1 It is a schematic flowchart of the intelligent risk monitoring method for deposit-free online lease customers in the embodiment of this application. Detailed Embodiment

[0048] The following combines the attached Figure 1 , and further elaborates on this application in detail.

[0049] The embodiment of this application discloses an intelligent risk monitoring method for deposit-free online lease customers.

[0050] An intelligent risk monitoring method for deposit-free online lease customers includes the following steps:

[0051] S1. Collect the historical lease user dataset, where the historical lease user dataset includes the personal data, external environment data, and corresponding overdue rent situations during the lease period of the user sample. After collecting the personal data and external environment data during the lease period of the historical lease users, clean the collected personal data and external environment data, remove data records with duplicates, errors, and excessive missing values, and discretize the continuous data according to the actual situation (for example, divide the income level into several intervals), etc.

[0052] For the dataset of online rental users without deposit, the number of samples of users who pay rent overdue is usually small and belongs to the minority class. Therefore, the following method can be used to generate new similar samples for the minority class of overdue rent samples to increase their proportion in the sample set and make the number of samples in each class relatively balanced. The specific steps are as follows:

[0053] Set a threshold for the proportion of the number of overdue rent samples in the total samples. When the proportion of the number of overdue rent samples in the collected historical rental user dataset is lower than the set threshold, for each overdue rent sample in the historical rental user dataset, find the k samples x i nearest to it in its feature space i1 ,x i2 ,...,x ik , randomly select one of the nearest neighbor samples x ij , and generate a new overdue rent sample x new by linear interpolation. The formula for generating the new overdue rent sample x new is as follows:

[0054] x new =x i +λ(x ij -x i ),

[0055] where λ is a random number between 0 and 1.

[0056] S2. Use the dataset of historical rental users to construct a decision tree model. In an optional embodiment, the construction of the decision tree model using the dataset of historical rental users includes:

[0057] Calculate the information entropy H(D) of the historical rental user dataset D through the following formula:

[0058]

[0059] where c is the number of classes (c takes 2, that is, the two situations of overdue and not overdue), and p i is the proportion of the i-th class of samples in the dataset D;

[0060] Calculate the conditional entropy H(D|A) of a certain feature A (such as the feature "income level") in the historical rental user dataset D through the following formula:

[0061]

[0062] where n is the number of values of the feature A (for example, if the income level is divided into three intervals: high, medium, and low, then n takes 3), D j represents the subset where the value of the feature A is j, and |D j| and | D | respectively represent the subset D j and the number of samples in the dataset D, H(D j ) is the subset D j 's information entropy;

[0063] Calculate the information gain Gain(D, A) of each feature in the historical rental user dataset D through the following formula:

[0064] Gain(D, A) = H(D) - H(D|A),

[0065] where H(D) is the information entropy of the historical rental user dataset D, and H(D|A) is the conditional entropy of a certain feature A in the historical rental user dataset D;

[0066] Select the feature with the largest information gain Gain(D, A) as the splitting feature of the current decision tree node, and build the decision tree accordingly. For example, after calculation, it is found that the feature "credit assessment" has the largest information gain, then use "credit assessment" to split the root node, and divide the dataset into different subsets according to its different values (such as good, medium, poor), and then repeat the above steps for each subset to continue building the decision tree until the stopping condition is met (such as all samples in the subset belong to the same category, or the number of samples in the subset is less than a certain set threshold, etc.).

[0067] S3. Collect the personal data and external environment data of new rental users during the machine rental period. In an optional embodiment, the personal data of historical rental users and new rental users during the machine rental period includes age, occupation, income situation, marital status, credit assessment data (such as credit scores and credit inquiry times on different platforms, etc.), rent payment behavior data (such as rent payment methods, sources of rent payment money, number of times and days of early or overdue rent payment, etc.), education level, residential stability, family financial status (family assets, family debt situation, and family major event situation, etc.), consumption behavior data (consumption structure, consumption frequency and amount, etc.), social network data (social activity and credit status of social circles), financial market data (holding situations of financial products such as stocks, funds, and bonds), and loan contract data (loan amount, loan term, and loan interest rate, etc.); the external environment data includes macroeconomic conditions (macroeconomic indicators such as GDP growth rate, unemployment rate, and inflation rate), industry trends, and regional difference situations.

[0068] In an optional embodiment, the personal data and external environment data of historical rental users and new rental users during the machine rental period can be collected through the following methods: cooperate with other relevant institutions to obtain the credit records and consumption behavior data of rental users on other platforms, etc.; collect public social network data and e-commerce platform consumption data, etc. through legal web crawling technology; design special questionnaires or encourage rental users to voluntarily provide more personal data through online customer service channels.

[0069] After collecting the personal data and external environment data of new lease users during the lease machine period, clean the collected personal data and external environment data, remove data records with excessive duplicates, errors, and missing values, and discretize continuous data according to the actual situation (for example, divide the income level into several intervals), etc.;

[0070] Calculate the overdue rent risk probability of new lease users through the constructed decision tree model. In an optional embodiment, calculating the overdue rent risk probability of new lease users through the constructed decision tree model includes:

[0071] For new lease users, starting from the root node of the decision tree constructed in step S2, follow the corresponding branches downward according to their corresponding feature values, and finally reach a certain leaf node. Calculate the overdue rent risk probability of the new lease user at the leaf node through the following formula

[0072]

[0073] where m is the number of users with overdue rent in the sample set included in the reached leaf node, and N is the total number of lease user samples at the leaf node.

[0074] When constructing a decision tree model using the dataset of historical lease users, due to factors such as insufficient dataset volume, data noise, or improper feature selection when constructing the decision tree, the constructed decision tree model may be unstable, thereby affecting the prediction performance and reliability of the model. Therefore, in an optional embodiment, the instability of the decision tree can be reduced through the following method, and the specific steps are as follows:

[0075] When constructing a decision tree model using the dataset of historical lease users, randomly sample with replacement from the historical lease user dataset D to obtain K sub-training datasets D 1 , D 2 ,... D K , where N is the number of samples in the historical lease user dataset D;

[0076] In each sub-training dataset D i , randomly select a feature subset F of size m from the entire feature set F i , where or m = log 2 |F|;

[0077] Use each sub-training dataset D i and the corresponding feature subset F i to train a decision tree T i respectively, to obtain a random forest model {T composed of K decision trees1 ,T 2 ,...,T k};

[0078] When calculating the overdue rent risk probability of new lease users, input it into each decision tree T i for prediction to obtain K prediction results and use the averaging method to obtain the overdue rent risk probability of new lease users

[0079] When constructing each decision tree through random forest, a sampling method of random sampling with replacement is used to extract samples from the original training dataset to form different sub-training datasets, so that the sample data on which each decision tree is based is different, thereby reducing the dependence of the decision tree on specific samples and reducing the possibility of large fluctuations in the decision tree structure and prediction results caused by small changes in the samples, and further reducing instability. When performing feature selection at each node of the decision tree construction, random forest does not use all features, but randomly selects a subset of features from all features, so that each decision tree learns and divides on different subsets of features, avoiding the decision tree relying too much on certain specific features, increasing the diversity between decision trees, and further reducing the instability of the decision tree. By integrating multiple different decision trees, synthesizing their prediction results, and using the averaging method to obtain the final prediction result, even if individual decision trees make incorrect or unstable predictions, they can be offset and corrected to a certain extent through integration, thereby improving the stability and accuracy of the overall model, and improving the prediction performance and reliability of the model.

[0080] Since a single decision tree model is used to predict the overdue rent risk probability of new lease users, the prediction result of the decision tree model may be affected by small changes in the data and the stability is relatively weak. Therefore, in an optional embodiment, the following method can be used to improve the prediction accuracy and stability, and the specific steps are as follows:

[0081] When constructing a decision tree model using the dataset of historical lease users, divide the historical lease user dataset into F 1 and F 2 two feature subsets, use the feature subset F 1 and the corresponding customer samples to train the decision tree model, use the feature subset F 2 and the corresponding customer samples to train the logistic regression model, respectively predict the overdue rent risk probability of new lease users through the decision tree model and the logistic regression model, and fuse the prediction results of the two models to obtain the fused overdue rent risk probability.

[0082] The hypothesis function of the logistic regression model is where hθ (x) is the output of the hypothesis function, where x is the customer feature vector based on the feature subset F 2 , θ is the parameter vector of the logistic regression model, e is the base of the natural logarithm, -θ T x is the inner product of the parameter vector θ and the feature vector x;

[0083] Estimate the parameter θ by the maximum likelihood estimation method to maximize the likelihood function of the logistic regression model on the training set. After training, for the new lease user sample x new , predict through the decision tree model and the logistic regression model respectively to obtain the overdue rent risk probability of the new lease user and

[0084] Fuse the prediction results of the two models to obtain the fused overdue rent risk probability, including: fuse the overdue rent risk probability of the new lease user according to the following formula and

[0085] where is the fused overdue rent risk probability, w 1 is the weight of the decision tree model, w 2 is the weight of the logistic regression model, and w 1 +w 2 = 1, is the overdue rent risk probability of the new lease user predicted by the decision tree model, is the overdue rent risk probability of the new lease user predicted by the logistic regression model.

[0086] By combining the decision tree model and the logistic regression model, the processing ability of the decision tree model for non-linear relationships and the processing ability of the logistic regression model for linear relationships can be fully utilized, as well as the feature selection ability of the decision tree model and the stability of the logistic regression model, thereby improving the prediction accuracy and stability of the overdue rent risk probability of the new lease user. Moreover, the combination of the two models can reduce the sensitivity of the model to data changes to a certain extent and improve the robustness of the model. When the data distribution changes or outliers appear, the complementary effects of the two models can better adapt to the changes and reduce the impact on the prediction results.

[0087] S4. Combine the overdue rent risk probability of new lease users, classify the new lease users into corresponding risk levels, and formulate corresponding preventive measures and collection strategies according to the risk levels of new lease users. For example: For users with a relatively low risk level of overdue rent risk probability, the rental platform can automatically send reminder notifications to users in various forms such as text messages, APP push notifications, or emails, reminding users to pay the rent in a timely manner. The notification content includes information such as the amount of rent to be paid and the due date. For users with a medium risk level of overdue rent risk probability, in addition to automatically sending reminder notifications to users through the rental platform, professional customer service staff can also be arranged to communicate with users by phone. The customer service staff can ask in a friendly and patient manner whether there are financial difficulties or other special circumstances, and determine whether the user can pay the rent on time and in full. If the user indicates that they cannot pay the rent on time and in full, an attempt can be made to negotiate a solution with the user, such as whether it is possible to pay the rent in installments. For users with a relatively high risk level of overdue rent risk probability, in addition to taking measures such as automatically sending reminder notifications to users and communicating with users by phone by customer service staff, staff can also be arranged to communicate with users face-to-face offline to determine the user's willingness to pay rent. If the user clearly indicates that they cannot pay the rent, a verbal or written warning can be issued to the user, informing the user of the possible legal consequences, and restricting some usage functions of the leased mobile phone.

[0088] To further improve the prediction accuracy and comprehensiveness of the overdue rent probability of new lease users, in an alternative embodiment, the prediction accuracy and comprehensiveness can also be improved through the following method. The specific steps are as follows:

[0089] During the period when a new lease user leases a mobile phone, collect the status data of the mobile phone. The status data includes the power-on duration P of the mobile phone usage , the abnormal location situation P location and the mobile phone repair situation P maintenance , and calculate the overdue rent probability P of the mobile phone status of the new lease user through the following formula phone :

[0090] P phone =β×P usage +γ×P location +δ×P maintenance ,

[0091] where β, γ, and δ are weight coefficients, and β + γ + δ = 1;

[0092] Calculate the overdue rent risk probability of new lease users through the constructed decision tree model Combine the overdue rent risk probability and the overdue rent probability P of the mobile phone status phone , and calculate the comprehensive overdue rent probability P through the following formula:

[0093] where α is a weight coefficient;

[0094] New rental users are classified into corresponding risk levels according to the comprehensive overdue rent probability P, and corresponding preventive measures and collection strategies are formulated according to the risk levels of new rental users.

[0095] The overdue rent behavior of rental users is often affected by the interaction of multiple factors, and the relationship between these factors may be very complex. Combining the user's mobile phone status and the risk probability of the user's overdue rent can better capture this complex risk relationship, and using the complementary information of the two dimensions can reduce the error and uncertainty of single-factor prediction, and further improve the prediction accuracy and comprehensiveness of the overdue rent probability of new rental users.

[0096] The embodiments of this specific implementation manner are all preferred embodiments of this application, and do not limit the protection scope of this application accordingly. The same components are denoted by the same reference numerals. Therefore, all equivalent changes made according to the structure, shape, and principle of this application should be covered within the protection scope of this application.

Claims

1. A method for intelligently monitoring customer risks of online rental services without deposit, characterized in that: include: Collecting a historical rental user data set, wherein the historical rental user data set includes personal data, external environment data, and corresponding overdue rental information of user samples during the rental period; Build a decision tree model using a dataset of historical rental users; Collect personal data and external environment data of new lease users during their lease period, and calculate the risk probability of overdue payment of new lease users through the constructed decision tree model; Based on the risk probability of overdue rent payment for new tenants, new tenants are divided into corresponding risk levels, and corresponding preventive measures and collection strategies are formulated according to the risk levels of new tenants.

2. According to claim 1, a method for intelligently monitoring customer risks of online rental services without deposit, characterized in that: The decision tree model is constructed using the data set of historical rental users, personal data and external environment data of new rental users during the rental period are collected, and the overdue payment risk probability of new rental users is calculated through the constructed decision tree model, including: The information entropy H(D) of the historical rental user dataset D is calculated by the following formula: Among them, c is the number of categories, p i is the proportion of samples of the i-th category in the data set D; The conditional entropy H(DA) of a feature A in the historical rental user dataset D is calculated using the following formula: Among them, n is the number of values ​​of feature A, D j represents the subset of feature A whose value is j, D j | and |D| represent subset D respectively. j and the number of samples in the dataset D, H(D j ) is a subset D j Information entropy of The information gain Gain(D,A) of each feature in the historical rental user dataset D is calculated by the following formula: Gain(D,A)=H(D)-H(DA), Among them, H(D) is the information entropy of the historical rental user data set D, and H(DA) is the conditional entropy of a feature A in the historical rental user data set D; Select the feature with the largest information gain Gain (D, A) as the partition feature of the current decision tree node to build a decision tree; For new tenants, start from the root node of the decision tree, go down along the corresponding branch according to its corresponding feature value, and finally reach a leaf node. The overdue rent risk probability of the new tenant at the leaf node is calculated by the following formula Among them, m is the number of overdue users in the sample set contained in the arrived leaf node, and N is the total number of rental user samples in the leaf node.

3. According to claim 1, a method for intelligently monitoring customer risks of online rental services without deposit, characterized in that: The personal data includes age, occupation, income, marital status, credit assessment data, rental behavior data, education level, residential stability, family financial status, consumer behavior data, social network data, financial market data and loan contract data; the external environment data includes macroeconomic conditions, industry trends and regional differences.

4. According to claim 1, a method for intelligently monitoring customer risks of online rental services without deposit, characterized in that: After collecting the personal data and external environment data of historical and new renters during their rental period, the collected personal data and external environment data are cleaned separately to remove data records with too many duplicates, errors, and missing values, and the continuous data is discretized according to actual conditions.

5. According to claim 1, a method for intelligently monitoring customer risks of online rental services without deposit, characterized in that: When using the historical rental user dataset to build a decision tree model, random sampling with replacement is performed from the historical rental user dataset D to obtain K sub-training datasets D1, D2, ...D of size N. K , where N is the number of samples in the historical rental user dataset D; Each sub-training dataset D i In the above example, a feature subset F of size m is randomly selected from the entire feature set F. i ,in Or m = log2|F|; Using each sub-training dataset D i and the corresponding feature subset F i Train a decision tree T respectively i , we get a random forest model consisting of K decision trees {T1,T2,...,T k }; When calculating the risk probability of overdue rent payment for new tenants, it is input into each decision tree T i Make predictions and get K prediction results The average method is used to obtain the risk probability of overdue rent payment for new tenants.

6. According to claim 1, a method for intelligently monitoring risks of online rental customers without deposit, characterized in that: Set a threshold for the proportion of overdue rent samples to the total samples. When the proportion of overdue rent samples in the collected historical rental user data set is lower than the set threshold, for each overdue rent sample in the historical rental user data set, in its feature space middle Find the sample x related to overdue rent i The k nearest neighbor samples x i1 ,x i2 ,...,x ik , randomly select a nearest neighbor sample x ij , and generate new overdue rent samples x by linear interpolation new , generate new overdue rent samples x new The formula is as follows: x new =x i +λ(x ij -x i ), Here, λ is a random number between 0 and 1.

7. According to claim 1, a method for intelligently monitoring customer risks of online rental services without deposit, characterized in that: When using the data set of historical rental users to build a decision tree model, the historical rental user data set is divided into two feature subsets, F1 and F2. The feature subset F1 and the corresponding customer samples are used to train the decision tree model, and the feature subset F2 and the corresponding customer samples are used to train the logistic regression model. The overdue rent risk probability of new rental users is predicted by the decision tree model and the logistic regression model respectively, and the prediction results of the two models are fused to obtain the fused overdue rent risk probability.

8. The method for intelligently monitoring risk of online rental customers without deposit according to claim 7 is characterized in that: The hypothesis function of the logistic regression model is Among them, h θ (x) is the output of the hypothesis function, where x is the customer feature vector based on feature subset F2, θ is the parameter vector of the logistic regression model, e is the base of the natural logarithm, -θ T x is the inner product of the parameter vector θ and the eigenvector x; The maximum likelihood estimation method is used to estimate the parameter θ, so that the likelihood function of the logistic regression model on the training set is maximized. After the training is completed, for the new rental user sample x new , respectively, using decision tree model and logistic regression model to predict the risk probability of overdue rent payment for new tenants and 9. The method for intelligently monitoring risk of online rental customers without deposit according to claim 8 is characterized in that: The prediction results of the two models are integrated to obtain the integrated overdue rent risk probability, including: integrating the overdue rent risk probability of the new rental user according to the following formula and in, is the risk probability of overdue rent payment after fusion, w1 is the weight of decision tree model, w2 is the weight of logistic regression model, and w1+w2=1, To predict the risk probability of overdue rent payment for new tenants through the decision tree model, To predict the risk probability of overdue rent payment for new tenants through a logistic regression model.

10. The method for intelligently monitoring customer risks of online rental services without deposit according to claim 1, characterized in that: During the rental period of the new user, the status data of the mobile phone is collected, and the status data includes the power-on time P of the mobile phone. usage 、Position abnormality P location and mobile phone repair situation maintenance , the probability P of overdue payment of mobile phone status of new renting users is calculated by the following formula phone : P phone =β×P usage +γ×P location +δ×P maintenance , Among them, β, γ, δ are weight coefficients, and β+γ+δ=1; The decision tree model is constructed to calculate the risk probability of overdue rent payment for new tenants Combined with the risk probability of overdue rent and the probability of overdue rent payment P of mobile phone status phone , and calculate the comprehensive overdue rent probability P by the following formula: Among them, α is the weight coefficient; New rental users are divided into corresponding risk levels according to the comprehensive overdue rent probability P.

Images