Model-based cost prediction method, device, computer equipment and storage medium

Abstract

The present application discloses a model-based cost prediction method, device, computer equipment and storage medium, which relates to the field of smart medical technology. By training sample data, a cost prediction model is constructed, and then the predicted cost output by the cost prediction model is obtained. Based on the predicted cost, a cost report is generated, so that the staff can communicate with the patient in advance based on the predicted cost, pay the secondary surgery fee, and close the case, saving a lot of manpower and material resources, simplifying procedures, and improving efficiency. The method includes: obtaining sample data, extracting features from the sample data based on a feature extraction algorithm, and obtaining multiple model features; using the Lightgbm algorithm to perform parameter conversion on multiple model features to obtain a cost prediction model; determining the data to be predicted, inputting the data to be predicted into the cost prediction model, and obtaining the predicted cost output by the cost prediction model; generating a cost report based on the predicted cost, and sending the cost report to the recipient.

Description

Technical Field

[0001] The present application relates to the field of smart healthcare, and in particular to a model-based cost prediction method, apparatus, computer equipment and storage medium. Background Art

[0002] With the rapid development of smart medical technology, the scope of claims involved in the insurance industry is becoming more and more extensive, and medical claims are a relatively common type of claims. Medical claims are a process in which claims personnel manually predict the claims costs based on the medical report issued by the hospital and the medical bills. Considering that sometimes after a traffic accident, the insured person needs a second surgery after a period of recovery, such as a second surgery to remove the steel plate or other internal fixation surgery. Therefore, in medical claims, claims personnel also need to predict the cost of the subsequent second surgery for the insured's reference.

[0003] In the related technologies, many smart medical platforms currently require claims personnel to trace back the case involved in the first surgery, reopen the case, collect claims materials such as treatment information and cost information again, conduct offline claims and related investigation procedures, and finally make medical cost predictions, payments or investigations for the second surgery when predicting the costs.

[0004] In the process of implementing the relevant technology, the applicant found that there are at least the following technical problems in the relevant technology:

[0005] Reopening a case involves repetitive work by claims adjusters, and requires them to communicate repeatedly with the policyholders and the hospital, which involves a lot of manpower and material resources, and the cost forecasting process is complicated and inefficient. Summary of the invention

[0006] In view of this, the present application provides a model-based cost forecasting method, apparatus, computer equipment and storage medium, the main purpose of which is to solve the current problem that the cost forecasting process involves a large amount of manpower and material resources, is complex and inefficient.

[0007] According to a first aspect of the present application, a model-based cost prediction method is provided, the method comprising:

[0008] Acquire sample data, and perform feature extraction on the sample data based on a feature extraction algorithm to obtain a plurality of model features, wherein the feature extraction algorithm is a filtering algorithm or a wrapping algorithm;

[0009] The Lightgbm algorithm is used to perform input conversion on the multiple model features to obtain a cost prediction model;

[0010] Determine the data to be predicted, input the data to be predicted into the cost prediction model, and obtain the predicted cost output by the cost prediction model;

[0011] Based on the predicted expenses, an expense report is generated and sent to a recipient.

[0012] Optionally, obtaining sample data includes:

[0013] Acquire multiple historical expense data, and query whether each of the multiple historical expense data has an associated preceding expense, wherein the preceding expense is reached earlier than the associated historical expense data and reaches the same patient;

[0014] Extracting a plurality of historical expense data of related preceding expenses as a plurality of candidate expense data, and filtering abnormal data from the plurality of candidate expense data;

[0015] Querying the round number marked on each candidate fee data in the plurality of candidate fee data, dividing the candidate fee data with the same round number into the same data group, calculating the total fee of each data group, and obtaining a plurality of total fees;

[0016] Based on the quartile method, the maximum total cost and the minimum total cost are eliminated from the multiple total cost to obtain the target total cost;

[0017] The candidate cost data included in the data group corresponding to the target cost total is used as the sample data.

[0018] Optionally, the feature extraction algorithm is used to extract features from the sample data to obtain multiple model features, including:

[0019] When the feature extraction algorithm is a filtering algorithm, the sample data is split according to the data items included in the sample data to obtain a plurality of sample data items and a project cost corresponding to each sample data item;

[0020] According to the feature extraction algorithm, statistics are collected on a plurality of project costs corresponding to each sample data item, and variances of the plurality of project costs corresponding to each sample data item are calculated to obtain a plurality of variances;

[0021] A target variance greater than a variance threshold is extracted from the plurality of variances, and sample data items for generating the target variance are used as the plurality of model features.

[0022] Optionally, the feature extraction algorithm is used to extract features from the sample data to obtain multiple model features, including:

[0023] When the feature extraction algorithm is a package algorithm, the sample data is split according to the data items included in the sample data to obtain multiple sample data items and the item cost corresponding to each sample data item;

[0024] Extracting sample data items from the plurality of sample data items to form a sample item set;

[0025] The package algorithm is used to extract the set features of the sample item set, a sample model is constructed based on the set features, test data is input into the sample model, and prediction result data output by the sample model is obtained;

[0026] Determine the actual result data corresponding to the test data, and calculate the data deviation between the actual result data and the predicted result data;

[0027] Obtaining a preset model accuracy value indicated by the package algorithm, comparing the data deviation with the preset model accuracy value, and determining an evaluation result of the sample model;

[0028] The sample item set is updated according to the evaluation result, and the sample model is reconstructed based on the updated sample item set. The retrained sample model is evaluated until the obtained evaluation result indicates that the trained sample model reaches the preset model accuracy value, and the sample data items included in the current sample item set are used as the multiple model features.

[0029] Optionally, the Lightgbm algorithm is used to perform input conversion on the multiple model features to obtain a cost prediction model, including:

[0030] The Lightgbm algorithm is used to construct the model and obtain multiple model parameters;

[0031] A training feature set is extracted from the multiple model features, and the model parameters are used to perform input parameter transformation on the training feature set to obtain the cost prediction model.

[0032] Optionally, after the Lightgbm algorithm is used to construct the model and multiple model parameters are obtained, the method further includes:

[0033] Extracting parameters to be adjusted from the multiple model parameters, and dividing the parameters to be adjusted into a preset number of parameter groups;

[0034] Determine a target parameter grouping, test the target parameter grouping with other parameter groups, and adjust parameters of the target parameter grouping according to the test results, wherein the target parameter grouping is any parameter grouping in the preset number of parameter groups, and the other parameter grouping is other parameter groupings in the preset number of parameter groups except the target parameter grouping;

[0035] A new target parameter grouping is re-determined from the preset number of parameter groups, and the new target parameter grouping is tested using new other parameter groups, and the new target parameter grouping is adjusted according to the test results until the adjustment of the preset number of parameter groups is completed.

[0036] Optionally, generating a cost report based on the predicted cost includes:

[0037] Parsing the preset fees, obtaining the predicted data items and the predicted data item fees included in the preset fees, listing the preset data items and the preset data item fees according to the corresponding relationship between the preset data items and the preset data item fees, and generating the fee report; or,

[0038] Acquire actual expenses, parse the actual expenses, obtain actual data items and actual data item expenses included in the actual expenses, compare the actual data items with the data item expenses corresponding to the preset data items, calculate the expense deviation, and mark the actual data items whose expense deviation is greater than the deviation threshold, and generate the expense report.

[0039] According to a second aspect of the present application, a model-based cost prediction device is provided, the device comprising:

[0040] An extraction module, used to obtain sample data, and perform feature extraction on the sample data based on a feature extraction algorithm to obtain a plurality of model features, wherein the feature extraction algorithm is a filtering algorithm or a wrapping algorithm;

[0041] A processing module, used for performing input parameter conversion on the plurality of model features by using the Lightgbm algorithm to obtain a cost prediction model;

[0042] An input module, used for determining data to be predicted, inputting the data to be predicted into the cost prediction model, and obtaining predicted costs output by the cost prediction model;

[0043] A generating module is used to generate an expense report based on the predicted expense and send the expense report to a recipient.

[0044] Optionally, the extraction module is used to obtain multiple historical cost data, query whether each of the multiple historical cost data has associated preceding costs, and the preceding costs are reached earlier than the associated historical cost data and reach the same patient; extract multiple historical cost data with associated preceding costs as multiple candidate cost data, and filter abnormal data among the multiple candidate cost data; query the round number marked by each candidate cost data in the multiple candidate cost data, divide the candidate cost data with the same round number into the same data group, calculate the total cost of each data group, and obtain multiple total cost; based on the quartile method, eliminate the maximum total cost and the minimum total cost from the multiple total cost to obtain a target total cost; and use the candidate cost data included in the data group corresponding to the target total cost as the sample data.

[0045] Optionally, the extraction module is used to, when the feature extraction algorithm is a filtering algorithm, split the sample data according to the data items included in the sample data to obtain multiple sample data items and project costs corresponding to each sample data item; according to the feature extraction algorithm, count the multiple project costs corresponding to each sample data item, calculate the variance of the multiple project costs corresponding to each sample data item, and obtain multiple variances; extract a target variance greater than a variance threshold from the multiple variances, and use the sample data item that generates the target variance as the multiple model features.

[0046] Optionally, the extraction module is used to, when the feature extraction algorithm is a package-type algorithm, split the sample data according to the data items included in the sample data to obtain multiple sample data items and the project cost corresponding to each sample data item; extract sample data items from the multiple sample data items to form a sample item set; use the package-type algorithm to extract the set features of the sample item set, build a sample model based on the set features, input test data into the sample model, and obtain prediction result data output by the sample model; determine the actual result data corresponding to the test data, and calculate the data deviation between the actual result data and the prediction result data; obtain the preset model accuracy value indicated by the package-type algorithm, compare the data deviation with the preset model accuracy value, and determine the evaluation result of the sample model; update the sample item set according to the evaluation result, and reconstruct the sample model based on the updated sample item set, and evaluate the retrained sample model until the obtained evaluation result indicates that the trained sample model reaches the preset model accuracy value, and the sample data items included in the current sample item set are used as the multiple model features.

[0047] Optionally, the processing module is used to use the Lightgbm algorithm to build a model to obtain multiple model parameters; extract a training feature set from the multiple model features, use the model parameters to perform input conversion on the training feature set, and obtain the cost prediction model.

[0048] Optionally, the processing module is also used to extract parameters to be adjusted from the multiple model parameters, and divide the parameters to be adjusted into a preset number of parameter groups; determine a target parameter group, use other parameter groups to test the target parameter group, and adjust parameters of the target parameter group according to the test results, the target parameter group is any parameter group in the preset number of parameter groups, and the other parameter group is other parameter groupings in the preset number of parameter groups except the target parameter group; re-determine a new target parameter group in the preset number of parameter groups, and use a new other parameter group to test the new target parameter group, and adjust the new target parameter group according to the test results, until the adjustment of the preset number of parameter groups is completed.

[0049] Optionally, the generation module is used to parse the preset fees, obtain the predicted data items and predicted data item fees included in the preset fees, list them according to the correspondence between the preset data items and the preset data item fees, and generate the fee report; or, obtain the actual fees, parse the actual fees, obtain the actual data items and actual data item fees included in the actual fees, compare the actual data items with the data item fees corresponding to the preset data items, calculate the fee deviation, and mark the actual data items whose fee deviation is greater than the deviation threshold to generate the fee report.

[0050] According to a third aspect of the present application, a computer device is provided, comprising a memory and a processor, wherein the memory stores a computer program, and the processor implements the steps of any one of the methods described in the first aspect when executing the computer program.

[0051] According to a fourth aspect of the present application, a storage medium is provided, on which a computer program is stored. When the computer program is executed by a processor, the steps of any one of the methods described in the first aspect are implemented.

[0052] Through the above technical scheme, the present application provides a model-based cost prediction method, device, computer equipment and storage medium. The present application obtains sample data, extracts features from the sample data based on a feature extraction algorithm, obtains multiple model features, and uses the Lightgbm algorithm to perform input parameter conversion on the multiple model features to obtain a cost prediction model, and then determines the data to be predicted, inputs the data to be predicted into the cost prediction model, obtains the predicted cost output by the cost prediction model, generates a cost report based on the predicted cost, sends the cost report to the recipient, and uses the feature extraction algorithm and the Lightgbm algorithm to construct a cost prediction model that can predict the cost of secondary surgery, so that staff can predict the cost of secondary surgery based on this cost prediction model, communicate with patients in advance, pay the cost of secondary surgery, and close the case, without the need for repeated communication between the personnel involved, saving a lot of manpower and material resources, simplifying the cost prediction procedure, and improving efficiency.

[0053] The above description is only an overview of the technical solution of the present application. In order to more clearly understand the technical means of the present application, it can be implemented in accordance with the contents of the specification. In order to make the above and other purposes, features and advantages of the present application more obvious and easy to understand, the specific implementation methods of the present application are listed below. BRIEF DESCRIPTION OF THE DRAWINGS

[0054] Various other advantages and benefits will become apparent to those of ordinary skill in the art by reading the detailed description of the preferred embodiments below. The accompanying drawings are only for the purpose of illustrating the preferred embodiments and are not to be considered as limiting the present application. Also, the same reference symbols are used throughout the accompanying drawings to represent the same components. In the accompanying drawings:

[0055] Figure 1 A schematic diagram of a cost prediction method based on a model provided in an embodiment of the present application is shown;

[0056] Figure 2 A schematic diagram of a cost prediction method based on a model provided in an embodiment of the present application is shown;

[0057] Figure 3 A schematic diagram of the structure of a model-based cost prediction device provided in an embodiment of the present application is shown;

[0058] Figure 4 A schematic diagram of the device structure of a computer device provided in an embodiment of the present application is shown. DETAILED DESCRIPTION

[0059] The exemplary embodiments of the present application will be described in more detail below with reference to the accompanying drawings. Although the exemplary embodiments of the present application are shown in the accompanying drawings, it should be understood that the present application can be implemented in various forms and should not be limited by the embodiments set forth herein. On the contrary, these embodiments are provided in order to enable a more thorough understanding of the present application and to fully convey the scope of the present application to those skilled in the art.

[0060] The present application embodiment provides a cost prediction method based on a model, such as Figure 1 As shown, the method includes:

[0061] 101. Obtain sample data, and perform feature extraction on the sample data based on a feature extraction algorithm to obtain multiple model features, wherein the feature extraction algorithm is a filtering algorithm or a wrapping algorithm.

[0062] 102. Use the Lightgbm algorithm to transform multiple model features into input parameters to obtain the cost prediction model.

[0063] 103. Determine the data to be predicted, input the data to be predicted into the cost prediction model, and obtain the predicted cost output by the cost prediction model.

[0064] 104. Generate an expense report based on the predicted expenses, and send the expense report to a recipient.

[0065] The method provided in the embodiment of the present application obtains sample data, performs feature extraction on the sample data based on a feature extraction algorithm, obtains multiple model features, and uses the Lightgbm algorithm to perform input parameter conversion on the multiple model features to obtain a cost prediction model, and then determines the data to be predicted, inputs the data to be predicted into the cost prediction model, obtains the predicted cost output by the cost prediction model, generates a cost report based on the predicted cost, sends the cost report to the recipient, and uses the feature extraction algorithm and the Lightgbm algorithm to construct a cost prediction model that can predict the cost of secondary surgery, so that the staff can predict the cost of secondary surgery based on this cost prediction model, communicate with the patient in advance, pay the cost of secondary surgery, and close the case, without the need for repeated communication between the personnel involved, saving a lot of manpower and material resources, simplifying the cost prediction process, and improving efficiency.

[0066] The present application embodiment provides a cost prediction method based on a model, such as Figure 2 As shown, the method includes:

[0067] 201. Obtain sample data.

[0068] At present, after a car accident, the patient will undergo surgery, and some patients will still need a second surgery some time after the surgery, such as a second internal fixation surgery such as removing the steel plate at a later time. The insurance company has closed the case when the patient had the first surgery. At this time, the patient applied for compensation again for the second surgery. The insurance company had to reopen the case, collect treatment information, cost information and other claim materials again, conduct offline claims and related investigation procedures, and finally make a second surgery medical expense compensation or investigation. However, the applicant recognized that this method has many defects: for the staff, the staff repeated their work, wasted a lot of manpower, material resources and financial resources, needed to communicate repeatedly with patients and hospitals, the procedures were complicated, the claims efficiency was low, and the fraudulent behavior of the second surgery costs of some patients could not be identified; for the patients, the patient experience is poor and the procedures are complicated.

[0069] Therefore, the present application proposes a cost prediction method, which can be applied to a smart medical platform, so that the smart medical platform can develop a cost prediction model for the cost of the secondary surgery based on historical secondary surgery cost data, using data analysis and machine learning technology. For patients who need a secondary surgery after the first surgery, the relevant data of the patient is input into the cost prediction model for prediction, so that the cost prediction model outputs the predicted cost of the secondary surgery. Based on this predicted cost, the staff communicates with the patient in advance. After the two parties reach an agreement, the cost of the secondary surgery is paid in advance and the case is closed. In addition, for cases applying for secondary surgery costs, anti-fraud auxiliary identification can also be performed based on the prediction model. When the actual cost of the case is higher than the maximum threshold of the predicted cost, the staff is prompted to focus on the investigation, and there is no need for the personnel involved to communicate repeatedly, saving a lot of manpower and material resources, simplifying the cost prediction process, and improving efficiency. Among them, the smart medical platform can be operated based on an independent server, or it can be operated based on a server that provides cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communications, middleware services, domain name services, security services, content delivery networks (Content Delivery Network, CDN), and basic cloud computing such as big data and artificial intelligence platforms. In this way, when a cost prediction model is subsequently built based on the sample, medical data as a sample can be downloaded from the server, such as personal health records, prescriptions, examination reports, etc.

[0070] In the embodiment of the present application, the cost prediction model constructed is to predict the cost of the secondary operation. Therefore, it is necessary to prepare data related to the historical secondary operation as sample data for model training. Among them, the sample data has associated pre-order costs, and the pre-order costs are reached earlier than the historical cost data associated with it and reach the same patient, that is, the sample data is the relevant data of the secondary operation, and the associated pre-order costs are also the cost data of the primary data. When obtaining sample data, first, obtain multiple historical cost data, query whether each historical cost data in multiple historical cost data has associated pre-order costs, and extract multiple historical cost data with associated pre-order costs as multiple candidate cost data. Specifically, the surgical claim data can be viewed in the historical claim data, and the target surgical claim data that has been operated twice is determined in the surgical claim data as the candidate cost data. Next, filter the abnormal data in the multiple candidate cost data, and query the round number marked by each candidate cost data in the multiple candidate cost data, divide the candidate cost data with the same round number into the same data group, calculate the total cost of each data group, and obtain multiple total costs. The purpose of filtering and dividing multiple candidate cost data is to reduce the types of data items in the sample data, integrate the sample data, and facilitate subsequent model training. For example, when filtering, if there are obvious errors in the name of the hospital visited in some candidate cost data, the candidate cost data with the wrong name of the hospital visited will be filtered. For example, when dividing, the operation is used as the execution round. Assuming that the round number of this statistics is "Orthopedic surgery of patient A on July 2, 2020", blood routine, urine routine, liver and kidney function, electrolytes, blood sugar, coagulation function and other examination items are required before orthopedic surgery. During the operation, steel plates and other items are required. After the operation, plaster external fixation and other items are required. Then, the candidate cost data corresponding to these items done by patient A on July 2, 2020 are divided into the same data group, and the total cost of the data group is used as the total cost of orthopedic surgery. Among them, it should be noted that, considering that multiple candidate cost data will involve a large number of operations of the same type, therefore, when counting the total cost, it is necessary to determine the name of the operation performed on the patient, and for the operation name, count the candidate cost data corresponding to the surgical items related to the operation name in the patient's relevant candidate cost data, and calculate the total cost of the surgical items as the total cost.

[0071] Then, considering that the cost of a certain surgery for most patients with common symptoms is basically the same, however, some patients in critical condition may spend more additional costs than patients with common symptoms for the same surgery, while patients in better condition may spend less for the same surgery than patients with common symptoms. Therefore, in order to avoid the impact of large and small surgical costs on the assessment of overall costs, when calculating the total cost of each round of execution, the maximum or minimum cost values ​​can be eliminated. Specifically, based on the quartile method, the maximum and minimum cost totals can be eliminated from multiple cost totals, and the multiple cost totals after elimination are used as sample data. The process of eliminating the maximum and minimum cost totals is as follows:

[0072] First, query the round name and injury part of the execution round corresponding to each total cost of the multiple total costs currently obtained, and divide the total costs with the same round name or the same injury part into the same cost group to obtain multiple cost groups. Then, use the quartile method to sort the total costs included in each cost group from small to large, determine the maximum total cost and the minimum total cost in the current cost group based on the median value, remove the maximum total cost and the minimum total cost in each total cost, obtain the target total cost, and use the candidate cost data included in the data group corresponding to the target total cost as sample data. For example, assuming that the costs included in the surgical group corresponding to the injury part of the arm are 15,890 yuan, 15,870 yuan, 14,670 yuan, and 16,890 yuan respectively, it can be determined that the maximum total cost in this group of surgical groups is 16,890 yuan and the minimum total cost is 14,670 yuan, so 16,890 yuan and 14,670 yuan are deleted from this group of surgical groups, and the candidate cost data included in the data group corresponding to 15,890 yuan and 15,870 yuan are used as sample data.

[0073] Through the above process, sample data is obtained, which may specifically include institution name, injury information, diagnosis information and basic information. The institution name may be the name of the institution where the patient is treated, such as the name of the hospital; the injury information may be relevant information about the part of the patient where the lesion or injury occurs, such as injury classification, injury site, etc.; the diagnosis information may be the content of the diagnosis certificate obtained by the patient at the institution where the patient is treated, such as the diagnosis name, diagnosis code, operation name, operation code, etc.; the basic information may be the patient's basic personal information, such as the patient's age, gender, etc.

[0074] 202. Based on the feature extraction algorithm, feature extraction is performed on the sample data to obtain multiple model features.

[0075] After the sample data is determined, feature extraction is started on the sample data to obtain model features for subsequent model training. The feature extraction algorithm can be a filtering algorithm, i.e., a filtering method (Filter), or a wrapping algorithm, i.e., a wrapping method (Wrapper). The following first describes the extraction process when the feature extraction algorithm is a filtering method:

[0076] First, the sample data is split according to the data items included in the sample data, and multiple sample data items and the project costs corresponding to each sample data item are obtained. Subsequently, according to the feature extraction algorithm, the multiple project costs corresponding to each sample data item are counted, and the variances of the multiple project costs corresponding to each sample data item are calculated to obtain multiple variances, and the target variance greater than the variance threshold is extracted from the multiple variances, and the sample data items that generate the target variance are used as multiple model features. In fact, the filtering method is to score each feature according to divergence or correlation, set a threshold or select the number of thresholds, and complete feature selection. When using the filtering method, if the variance is used for filtering, the variance of all data corresponding to each sample data item is calculated. The larger the variance, the greater the difference of the sample data item, which is suitable for classification. If the variance is small, it proves that the difference within the sample data item is small and is not suitable for classification. When the variance is 0, the feature values ​​are the same at this time. If the F test is used, the F statistic (between-group variance / within-group variance) needs to be calculated, and then the F table is checked. If it is greater than the critical value (generally at a significance level of 0.05), the null hypothesis is rejected, that is, there is a significant difference between the groups. Furthermore, the correlation can be calculated by calculating the "estimate of the mutual information between each data item and its related data items". This estimate takes values ​​between [0, 1], where 0 means that the two variables are independent and 1 means that the two variables are completely correlated.

[0077] Next, the extraction process of the feature extraction algorithm using the wrapping method is described:

[0078] First, the sample data is split according to the data items included in the sample data, and multiple sample data items and the project costs corresponding to each sample data item are obtained. Subsequently, sample data items are extracted from multiple sample data items to form a sample item set, and the package algorithm is used to extract the set features of the sample item set. A sample model is constructed based on the set features, and the test data is input into the sample model to obtain the prediction result data output by the sample model. Next, the actual result data corresponding to the test data is determined, the data deviation between the actual result data and the prediction result data is calculated, the preset model accuracy value indicated by the package algorithm is obtained, the data deviation is compared with the preset model accuracy value, and the evaluation result of the sample model is determined. The sample item set is updated according to the evaluation result, and the sample model is rebuilt based on the updated sample item set. The retrained sample model is evaluated until the evaluation result indicates that the trained sample model reaches the preset model accuracy value, and the sample data items included in the current sample item set are used as multiple model features.

[0079] In fact, the wrapping method is to select a specific algorithm, and then select a feature set based on the algorithm effect (that is, the preset model accuracy value), select or delete some features, and see whether the algorithm model accuracy meets the standard, so as to confirm the features. Among them, the wrapping method is a method of simultaneous feature selection and algorithm training, which relies on the choice of the algorithm itself, such as coef_attribute or feature_importances_attribute to complete feature selection. For each set of sample items to be selected, the model is trained on the training set, and then the sample item set is selected on the test set according to the error size. The premise is to select the classification (or regression) algorithm to be used later, such as Random (random number algorithm), Forest (random forest), SVM (Support Vector Machine), KNN (k-Nearest Neighbor, nearest neighbor classification algorithm), etc. Specifically, the selected specific algorithm can be forward search, backward search and recursive feature elimination method: forward search, and initially empty, each time adding one to the feature set until K. Backward search, that is, all features are initially, and each time one is reduced and excluded from the feature set until K is left. Recursive feature elimination, a greedy optimization algorithm, finds the best performing sample item set. It repeatedly creates models and retains the best features or removes the worst features in each iteration (operating on N features). In the next iteration, it uses the features that were not selected in the previous modeling to build the next model until all features are exhausted. Then, it ranks the features according to the order in which it retains or removes them, and finally selects the best sample item set as the model features.

[0080] 203. The Lightgbm algorithm is used to build the model and obtain multiple model parameters.

[0081] Among them, after the model features are determined, the training of the model parameters is started, and then the determined model features are trained with the model parameters obtained by training to obtain the cost prediction model. Among them, this application adopts the Lightgbm (LightGradient Boosting Machine, gradient boosting) algorithm to construct the model parameters, and the specific process is as follows: First, set the initial parameters. When setting the initial parameters, set the boosting type to GBDT, the objective function to regression, the evaluation function to rmse, and the learning rate to 0.1. Subsequently, the set initial parameters are trained based on the Lightgbm algorithm to obtain the model parameters of the prediction model. The model parameters specifically include parameters such as num_leaves, max_depth, min_data_in_leaf, and feature_fraction. Next, considering that some of the model parameters obtained by training have poor values, it is easy to cause errors in the prediction process of the subsequent prediction model. Therefore, in this application, the model parameters will be adjusted. The specific process is as follows:

[0082] First, extract the parameters to be adjusted from multiple model parameters, divide the parameters to be adjusted into a preset number of parameter groups, determine the target parameter group, use other parameter groups to test the target parameter group, and adjust the parameters of the target parameter group according to the test results, wherein the target parameter group is any parameter group in the preset number of parameter groups, and the other parameter groups are other parameter groups in the preset number of parameter groups except the target parameter group. Then, re-determine a new target parameter group in the preset number of parameter groups, and use new other parameter groups to test the new target parameter group, and adjust the new target parameter group according to the test results, until the adjustment of the preset number of parameter groups is completed.

[0083] In the actual application process, the above model parameter adjustment process can be implemented using the (5-fold) validation method, that is, the cross (5-fold) validation method is used to tune parameters such as num_leaves, max_depth, min_data_in_leaf, feature_fraction, etc. Specifically, when using the 5-fold cross validation method, the model parameters to be adjusted can be divided into 5 groups, one of which is selected as the test set (that is, the target parameter grouping), and the other four groups are used as training sets (that is, other parameter groups). The training set is tested based on the test set, and the parameters of the training set are adjusted according to the test results. After the adjustment is completed, reselect the test set and retest and adjust until each group has been selected as the test set.

[0084] It should be noted that when adjusting parameters, usually num_leaves = 2^(max_depth), that is, the value of num_leaves should be set to be less than 2^(max_depth), otherwise it may cause overfitting; min_data_in_leaf needs to be appropriately increased. Increasing its value can prevent overfitting, and its value is usually set to a relatively large value; feature_fraction is used to set the proportion of features used in each iteration. In addition, the model parameters also involve the max_bin parameter. Reducing the value of max_bin can increase the model training speed, and increasing its value has a similar effect to increasing num_leaves; it also involves the bagging_fraction parameter, which can make bagging run faster and produce results. This application does not specifically limit which model parameters to adjust and how to adjust the model parameters.

[0085] 204. Extract a training feature set from multiple model features, use model parameters to perform input conversion on the training feature set, and obtain a cost prediction model.

[0086] Among them, after determining the model features, in order to be able to evaluate the prediction effect of the constructed cost prediction model so as to adjust the cost prediction model later, this application will split the model features, split the model features into a training feature set and a test feature set, and construct the cost prediction model based on the training feature set, and train and evaluate the constructed cost prediction model based on the test feature set, and then evaluate the accuracy of the cost prediction model by comparing the rmse values ​​of the results output by the cost prediction model and the evaluation model. Specifically, when splitting the training feature set and the test feature set, the model features can be randomly split according to a ratio of 8:2 to obtain the training feature set and the test feature set.

[0087] Next, the training feature set is transformed into an input parameter using the model parameters to obtain the cost prediction model. In addition, since the model features with a ratio of 2 are used as the test set in this application, the test set can be processed using the model parameters to obtain the evaluation model, and the RMSE values ​​of the results of the cost prediction model and the evaluation model output based on the same data are compared to evaluate the accuracy of the cost prediction model.

[0088] It should be noted that after the cost prediction model is generated, the cost prediction model can be saved. In the actual application process, the cost prediction model may be applied to a certain function of a certain platform, such as the cost prediction function of an insurance platform or a smart medical platform, so that the cost prediction model automatically outputs the predicted cost to the staff after the staff operates online. Therefore, the cost prediction model can be packaged and configured on the target platform, and a prediction entry is provided on the front end of the target platform. When the user triggers the prediction entry, the data to be predicted input by the user after the trigger is obtained, and the data to be predicted is input into the cost prediction model, so that the cost prediction model outputs the predicted cost based on the data to be predicted. This application does not specifically limit the application scenario of the cost prediction model.

[0089] 205. Determine the data to be predicted, input the data to be predicted into the cost prediction model, and obtain the predicted cost output by the cost prediction model.

[0090] In the embodiment of the present application, after the cost prediction model is established through the above process, the cost of the secondary surgery can be predicted based on the cost prediction model. Specifically, it is necessary to obtain the data to be predicted of the patient to be predicted, input the data to be predicted into the cost prediction model, and obtain the predicted cost of the secondary surgery output by the cost prediction model. It should be noted that since the cost prediction model constructed above uses the institution name, injury information, diagnosis information and basic information such as injury site, diagnosis name, operation name, hospital name, hospital level, whether to sue, and age of the injured as model features for training, therefore, when obtaining the patient's data to be predicted, the patient's injury site, diagnosis name, operation name, hospital name, hospital level, whether to sue, and age of the injured can also be used as the data to be predicted to ensure that the data to be predicted meets the content that the cost prediction model needs to be based on when making predictions, so that the prediction is more accurate.

[0091] 206. Generate expense reports based on predicted expenses.

[0092] In the embodiment of the present application, after the predicted cost of the second surgery is obtained, a cost report can be generated based on the predicted cost of the second surgery. Specific cost reports can be divided into two types, one is a cost detail report, which lists the data items involved in the predicted cost and the predicted data item costs, so that subsequent claims personnel can inform the patient of which items may be performed during the second surgery and the total cost of the items. The generation process is as follows: parse the preset costs, obtain the predicted data items and predicted data item costs included in the preset costs, list them according to the corresponding relationship between the preset data items and the preset data item costs, and generate a cost report.

[0093] The other is a cost comparison report, which is applicable to situations where a secondary surgery claim has been completed. It lists the actual situation of the secondary surgery claim and the data items involved in the predicted secondary surgery cost, calculates the deviation of each data item, and then evaluates which data item's claim is higher than the predicted cost deviation threshold according to the deviation, prompting the claims adjuster that there is a risk, and the claims adjuster conducts further verification. The generation process is as follows: obtain the actual cost, parse the actual cost, obtain the actual data items included in the actual cost and the actual data item costs, compare the actual data items with the data item costs corresponding to the preset data items, calculate the cost deviation, and mark the actual data items whose cost deviation is greater than the deviation threshold to generate a cost report.

[0094] In this way, this application uses the cost prediction model to predict the medical expenses of the second surgery in advance for patients who need a second surgery after the first surgery. Based on this, the staff communicates with the patient in advance, and after the two parties reach an agreement, the second surgery fee is paid in advance and the case is closed, which greatly saves operating and claims costs, improves claims efficiency and customer satisfaction. In addition, for cases applying for second surgery fees, anti-fraud auxiliary identification can be performed. When the actual cost is higher than the predicted cost deviation threshold, the claims personnel are prompted to focus on the investigation.

[0095] 207. Send the expense report to the recipient.

[0096] In the embodiment of the present application, after the expense report is generated, the smart medical platform will send the expense report to the recipient. The recipient can be a staff member or a patient. Specifically, the platform can send the expense report to the account registered by the staff member or patient in the platform, or it can also receive the email address, mobile phone number, etc. entered by the staff member or patient in the platform, and send the expense report to the email address or mobile phone. The present application does not specifically limit the transmission method of the expense report.

[0097] The method provided in the embodiment of the present application obtains sample data, performs feature extraction on the sample data based on a feature extraction algorithm, obtains multiple model features, and uses the Lightgbm algorithm to perform input parameter conversion on the multiple model features to obtain a cost prediction model, and then determines the data to be predicted, inputs the data to be predicted into the cost prediction model, obtains the predicted cost output by the cost prediction model, generates a cost report based on the predicted cost, sends the cost report to the recipient, and uses the feature extraction algorithm and the Lightgbm algorithm to construct a cost prediction model that can predict the cost of secondary surgery, so that the staff can predict the cost of secondary surgery based on this cost prediction model, communicate with the patient in advance, pay the cost of secondary surgery, and close the case, without the need for repeated communication between the personnel involved, saving a lot of manpower and material resources, simplifying the cost prediction process, and improving efficiency.

[0098] Further, as Figure 1 The specific implementation of the method, the embodiment of the present application provides a cost prediction device based on a model, such as Figure 3 As shown, the device includes: an extraction module 301, a processing module 302, an input module 303 and a generation module 304.

[0099] The extraction module 301 is used to obtain sample data, and perform feature extraction on the sample data based on a feature extraction algorithm to obtain multiple model features, wherein the feature extraction algorithm is a filtering algorithm or a wrapping algorithm;

[0100] The processing module 302 is used to use the Lightgbm algorithm to perform input conversion on the multiple model features to obtain a cost prediction model;

[0101] The input module 303 is used to determine the data to be predicted, input the data to be predicted into the cost prediction model, and obtain the predicted cost output by the cost prediction model;

[0102] The generating module 304 is used to generate an expense report based on the predicted expense and send the expense report to a recipient.

[0103] In a specific application scenario, the extraction module 301 is used to obtain multiple historical cost data, query whether each of the multiple historical cost data has associated preceding costs, and the preceding costs are reached earlier than the associated historical cost data and reach the same patient; extract multiple historical cost data with associated preceding costs as multiple candidate cost data, and filter abnormal data among the multiple candidate cost data; query the round number marked by each candidate cost data in the multiple candidate cost data, divide the candidate cost data with the same round number into the same data group, calculate the total cost of each data group, and obtain multiple total cost; based on the quartile method, eliminate the maximum total cost and the minimum total cost from the multiple total cost to obtain the target total cost; and use the candidate cost data included in the data group corresponding to the target total cost as the sample data.

[0104] In a specific application scenario, the extraction module 301 is used to split the sample data according to the data items included in the sample data when the feature extraction algorithm is a filtering algorithm, to obtain multiple sample data items and the project cost corresponding to each sample data item; according to the feature extraction algorithm, count the multiple project costs corresponding to each sample data item, calculate the variance of the multiple project costs corresponding to each sample data item, and obtain multiple variances; extract a target variance greater than a variance threshold from the multiple variances, and use the sample data item that generates the target variance as the multiple model features.

[0105] In a specific application scenario, the extraction module 301 is used to, when the feature extraction algorithm is a package algorithm, split the sample data according to the data items included in the sample data to obtain multiple sample data items and the project cost corresponding to each sample data item; extract sample data items from the multiple sample data items to form a sample item set; use the package algorithm to extract the set features of the sample item set, build a sample model based on the set features, input the test data into the sample model, and obtain the prediction result data output by the sample model; determine the actual result data corresponding to the test data, and calculate the data deviation between the actual result data and the prediction result data; obtain the preset model accuracy value indicated by the package algorithm, compare the data deviation with the preset model accuracy value, and determine the evaluation result of the sample model; update the sample item set according to the evaluation result, and reconstruct the sample model based on the updated sample item set, and evaluate the retrained sample model until the obtained evaluation result indicates that the trained sample model reaches the preset model accuracy value, and the sample data items included in the current sample item set are used as the multiple model features.

[0106] In a specific application scenario, the processing module 302 is used to use the Lightgbm algorithm to build a model to obtain multiple model parameters; extract a training feature set from the multiple model features, use the model parameters to perform input conversion on the training feature set, and obtain the cost prediction model.

[0107] In a specific application scenario, the processing module 302 is also used to extract the parameters to be adjusted from the multiple model parameters, and divide the parameters to be adjusted into a preset number of parameter groups; determine a target parameter group, use other parameter groups to test the target parameter group, and adjust the parameters of the target parameter group according to the test results, the target parameter group is any parameter group in the preset number of parameter groups, and the other parameter group is other parameter groupings in the preset number of parameter groups except the target parameter group; re-determine a new target parameter group in the preset number of parameter groups, and use a new other parameter group to test the new target parameter group, and adjust the new target parameter group according to the test results until the adjustment of the preset number of parameter groups is completed.

[0108] In a specific application scenario, the generation module 304 is used to parse the preset fees, obtain the predicted data items and predicted data item fees included in the preset fees, list them according to the correspondence between the preset data items and the preset data item fees, and generate the fee report; or, obtain the actual fees, parse the actual fees, obtain the actual data items and actual data item fees included in the actual fees, compare the actual data items with the data item fees corresponding to the preset data items, calculate the fee deviation, and mark the actual data items whose fee deviation is greater than the deviation threshold to generate the fee report.

[0109] The device provided in the embodiment of the present application obtains sample data, performs feature extraction on the sample data based on a feature extraction algorithm, obtains multiple model features, and uses the Lightgbm algorithm to perform input parameter conversion on the multiple model features to obtain a cost prediction model, and then determines the data to be predicted, inputs the data to be predicted into the cost prediction model, obtains the predicted cost output by the cost prediction model, generates a cost report based on the predicted cost, sends the cost report to the recipient, and uses the feature extraction algorithm and the Lightgbm algorithm to construct a cost prediction model that can predict the cost of secondary surgery, so that the staff can predict the cost of secondary surgery based on this cost prediction model, communicate with the patient in advance, pay the cost of secondary surgery, and close the case, without the need for repeated communication between the personnel involved, saving a lot of manpower and material resources, simplifying the cost prediction process, and improving efficiency.

[0110] It should be noted that for other corresponding descriptions of the functional units involved in the model-based cost prediction device provided in the embodiment of the present application, reference can be made to Figure 1 and Figure 2 The corresponding description is given in , which will not be repeated here.

[0111] In an exemplary embodiment, see Figure 4 , and also provides a computer device, which includes a bus, a processor, a memory and a communication interface, and may also include an input and output interface and a display device, wherein each functional unit can communicate with each other through the bus. The memory stores a computer program, and the processor is used to execute the program stored in the memory and execute the model-based cost prediction method in the above embodiment.

[0112] A computer-readable storage medium stores a computer program, which implements the steps of the model-based cost prediction method when executed by a processor.

[0113] Through the description of the above implementation mode, those skilled in the art can clearly understand that the present application can be implemented by hardware, or by means of software plus necessary general hardware platforms. Based on such understanding, the technical solution of the present application can be embodied in the form of a software product, which can be stored in a non-volatile storage medium (which can be a CD-ROM, a USB flash drive, a mobile hard disk, etc.), including several instructions to enable a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in each implementation scenario of the present application. Those skilled in the art can understand that the accompanying drawings are only schematic diagrams of a preferred implementation scenario, and the modules or processes in the accompanying drawings are not necessarily necessary for implementing the present application. Those skilled in the art can understand that the modules in the devices in the implementation scenario can be distributed in the devices of the implementation scenario according to the description of the implementation scenario, or can be changed accordingly and located in one or more devices different from the present implementation scenario. The modules of the above implementation scenarios can be combined into one module, or can be further split into multiple submodules. The above application serial numbers are only for description and do not represent the advantages and disadvantages of the implementation scenarios. The above disclosure is only a few specific implementation scenarios of the present application, but the present application is not limited to this, and any changes that can be thought of by those skilled in the art should fall within the scope of protection of the present application.

Claims

1. A cost prediction method based on a model, characterized in that: include: Obtain sample data, and perform feature extraction on the sample data based on a feature extraction algorithm to obtain multiple model features, wherein the feature extraction algorithm is a filtering algorithm or a wrapping algorithm; wherein obtaining sample data includes: obtaining multiple historical cost data, querying whether each of the multiple historical cost data has associated preceding costs, and the preceding costs are reached earlier than the associated historical cost data and reach the same patient; extracting multiple historical cost data with associated preceding costs as multiple candidate cost data, and filtering out abnormal data in the multiple candidate cost data; querying the round number marked on each candidate cost data in the multiple candidate cost data, and filtering out the abnormal data in the multiple candidate cost data; The candidate cost data with the same round number are divided into the same data group, and the total cost of each data group is calculated to obtain multiple total cost; based on the quartile method, the maximum total cost and the minimum total cost are eliminated from the multiple total cost to obtain the target total cost; the candidate cost data included in the data group corresponding to the target total cost is used as the sample data; wherein, when the feature extraction algorithm is a filtering algorithm, the sample data is split according to the data items included in the sample data to obtain multiple sample data items and the project cost corresponding to each sample data item; according to the feature extraction algorithm, the multiple project costs corresponding to each sample data item are counted, and the cost of each sample data item is calculated. The variances of the multiple project costs corresponding to the data items are calculated to obtain multiple variances, a target variance greater than a variance threshold is extracted from the multiple variances, and the sample data items that generate the target variance are used as the multiple model features; when the feature extraction algorithm is a parcel algorithm, the sample data is split according to the data items included in the sample data to obtain multiple sample data items and the project costs corresponding to each sample data item, sample data items are extracted from the multiple sample data items to form a sample item set, the parcel algorithm is used to extract the set features of the sample item set, a sample model is constructed based on the set features, and the test data is input into the sample model to obtain the sample item set. The predicted result data output by the model is used to determine the actual result data corresponding to the test data, calculate the data deviation between the actual result data and the predicted result data, obtain the preset model accuracy value indicated by the package algorithm, compare the data deviation with the preset model accuracy value, determine the evaluation result of the sample model, update the sample item set according to the evaluation result, and rebuild the sample model based on the updated sample item set, evaluate the retrained sample model, until the obtained evaluation result indicates that the trained sample model reaches the preset model accuracy value, and use the sample data items included in the current sample item set as the multiple model features; The Lightgbm algorithm is used to perform input conversion on the multiple model features to obtain a cost prediction model; Determine the data to be predicted, input the data to be predicted into the cost prediction model, and obtain the predicted cost output by the cost prediction model; Based on the predicted expenses, an expense report is generated and sent to a recipient.

2. The method according to claim 1, characterized in that The Lightgbm algorithm is used to perform parameter conversion on the multiple model features to obtain a cost prediction model, including: The Lightgbm algorithm is used to construct the model and obtain multiple model parameters; A training feature set is extracted from the multiple model features, and the model parameters are used to perform input parameter transformation on the training feature set to obtain the cost prediction model.

3. The method according to claim 2, characterized in that After the model is constructed using the Lightgbm algorithm and multiple model parameters are obtained, the method further includes: Extracting parameters to be adjusted from the multiple model parameters, and dividing the parameters to be adjusted into a preset number of parameter groups; Determine a target parameter grouping, test the target parameter grouping with other parameter groups, and adjust parameters of the target parameter grouping according to the test results, wherein the target parameter grouping is any parameter grouping in the preset number of parameter groups, and the other parameter grouping is other parameter groupings in the preset number of parameter groups except the target parameter grouping; A new target parameter grouping is re-determined from the preset number of parameter groups, and the new target parameter grouping is tested using new other parameter groups, and the new target parameter grouping is adjusted according to the test results until the adjustment of the preset number of parameter groups is completed.

4. The method according to claim 1, characterized in that The generating of a cost report based on the predicted cost includes: parsing the predicted cost to obtain the predicted data items and predicted data item costs included in the predicted cost, listing the predicted data items and the predicted data item costs according to the corresponding relationship between the predicted data items and the predicted data item costs, and generating the cost report; or, Acquire actual expenses, parse the actual expenses, obtain actual data items and actual data item expenses included in the actual expenses, compare the actual data items with the data item expenses corresponding to the predicted data items, calculate the expense deviation, and mark the actual data items whose expense deviation is greater than the deviation threshold, and generate the expense report.

5. A cost prediction device based on a model, characterized in that: include: An extraction module is used to obtain sample data, and based on a feature extraction algorithm, perform feature extraction on the sample data to obtain multiple model features, wherein the feature extraction algorithm is a filtering algorithm or a wrapping algorithm; the extraction module is used to obtain multiple historical cost data, query whether each of the multiple historical cost data has an associated preceding cost, and the preceding cost is reached earlier than its associated historical cost data and reaches the same patient; extract multiple historical cost data with associated preceding costs as multiple candidate cost data, and filter out abnormal data in the multiple candidate cost data; query the round number marked on each candidate cost data in the multiple candidate cost data, and filter out the round number on each candidate cost data. The candidate cost data with the same label are divided into the same data group, and the total cost of each data group is calculated to obtain multiple cost totals; based on the quartile method, the maximum cost total and the minimum cost total are eliminated from the multiple cost totals to obtain the target cost total; the candidate cost data included in the data group corresponding to the target cost total is used as the sample data; wherein, the extraction module is used for, when the feature extraction algorithm is a filtering algorithm, splitting the sample data according to the data items included in the sample data to obtain multiple sample data items and the project cost corresponding to each sample data item, and according to the feature extraction algorithm, counting the multiple project costs corresponding to each sample data item, and calculating each The variances of the multiple project costs corresponding to the sample data items are obtained to obtain multiple variances, a target variance greater than a variance threshold is extracted from the multiple variances, and the sample data items that generate the target variance are used as the multiple model features; the extraction module is used to split the sample data according to the data items included in the sample data when the feature extraction algorithm is a parcel-type algorithm, to obtain multiple sample data items and the project costs corresponding to each sample data item, to extract sample data items from the multiple sample data items to form a sample item set, to extract the set features of the sample item set using the parcel-type algorithm, to construct a sample model based on the set features, and to input test data into the sample model. Obtaining the prediction result data output by the sample model, determining the actual result data corresponding to the test data, calculating the data deviation between the actual result data and the prediction result data, obtaining the preset model accuracy value indicated by the package algorithm, comparing the data deviation with the preset model accuracy value, determining the evaluation result of the sample model, updating the sample item set according to the evaluation result, and reconstructing the sample model based on the updated sample item set, evaluating the retrained sample model until the obtained evaluation result indicates that the trained sample model reaches the preset model accuracy value, and using the sample data items included in the current sample item set as the multiple model features; A processing module, used for performing input parameter conversion on the plurality of model features by using the Lightgbm algorithm to obtain a cost prediction model; An input module, used for determining data to be predicted, inputting the data to be predicted into the cost prediction model, and obtaining predicted costs output by the cost prediction model; A generating module is used to generate an expense report based on the predicted expense and send the expense report to a recipient.

6. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 4 are implemented.

7. A storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 4 are implemented.

Images