WO2019169826A1 - Risk control method for determining irregular medical insurance behavior by means of data analysis - Google Patents

Abstract

The present application relates to a risk control method for determining irregular medical insurance behavior by means of data analysis. The method comprises: step 1: acquiring current and historical medical treatment behavior data of a person under risk control and personal information and public data related to the person under risk control; step 2: extracting, from the data acquired in step 1, features related to medical treatment behavior of the person under risk control; step 3: classifying, according to the extracted features, each medical treatment behavior of the current and historical medical treatment behavior of the person under risk control into corresponding medical treatment categories, so as to generate a medical treatment category sequence; step 4: using the medical treatment category sequence generated in step 3 as an observation sequence, using a medical treatment purpose as a hidden state, and calculating, according to a Hidden Markov model, a most likely hidden state sequence which comprises a most likely medical treatment purpose; and step 5: if the most likely medical treatment purpose of the person under risk control comprised in the hidden state sequence corresponds to an abnormal purpose, outputting medical data related to the current and historical medical treatment behavior of the person under risk control.

Description

Wind control method for inferring medical insurance violations through data analysis

This application claims the priority of the Chinese patent application filed on March 8, 2018, the application number of 201101191862.5, entitled "Introduction of the risk of medical insurance violations through data analysis", the overall content of which is referenced. The way is combined in this application.

Technical field

The present application relates to the field of Internet data processing technologies, and in particular, to a risk control method for inferring medical insurance violations through data analysis, providing a basis for system auditing or manual auditing.

Background technique

In the medical insurance social system, there are tens of thousands of transaction data of outpatient and hospital medical behaviors every day, mainly including transactions between patients and medical institutions, and transactions between medical institutions and insurance institutions. At present, the existing medical insurance processing system is difficult to accurately identify the real needs of patients when dealing with payment transactions, and the insured person or medical institution has the possibility of seeking improper benefits. The existence of medical fraud has seriously affected the balance of income and expenditure of medical insurance funds, infringing the interests of insured persons and social welfare.

The government and relevant departments have been committed to using big data methods to identify health insurance fraud and control health care risks. However, the existing medical insurance risk control scheme is mostly based on setting the threshold red line to monitor fraud, waste, abuse and other irregularities. Because the violations often change with the medical insurance policy, payment methods and supervision, the simple threshold division is not suitable. The actual application environment of multiple scenarios, insured personnel, and policy replacement.

Summary of the invention

In view of this, in view of the above disadvantages of the prior art, there is a need to solve the above problems by using technical means such as machine learning and classifier modeling.

According to an embodiment of the present application, there is provided a risk control method for inferring a medical insurance violation behavior by data analysis, characterized in that the purpose of the medical treatment object is divided into normal purpose and abnormal purpose, and the method comprises the following steps: Step 1. Obtain current and historical medical behavior data of the wind control object, and personal information and public data related to the wind control object; Step 2. Extract features related to the visiting behavior of the wind control object from the data acquired in step 1 Step 3: According to the extracted features, each visit behavior in the current and historical visiting behaviors of the wind control object is divided into corresponding medical treatment categories to form a medical treatment category sequence; Step 4, the medical treatment category formed in step 3 The sequence is used as an observation sequence, and the purpose of the treatment is regarded as an implicit state, and the most likely sequence of hidden states is calculated according to the hidden Markov model; step 5, if one or more hidden in the sequence of the most likely implicit state is included The containing state corresponds to the abnormal purpose, and outputs medical data related to the current and historical visiting behavior of the wind control object.

According to an embodiment of the present application, there is provided a wind control system for performing the foregoing method, comprising: a wind control object data acquisition module configured to acquire current and historical medical behavior data of the wind control object, and the wind control Object-related personal information and public data; a feature extraction module configured to extract features related to the visiting behavior of the wind control object from the data acquired by the wind control object data acquiring module; the medical treatment classification module is configured to be used for According to the extracted features, each visit behavior in the current and historical visiting behavior of the wind control object is divided into corresponding medical treatment categories to form a medical treatment category sequence; a diagnosis target estimation module is configured to be used in the medical treatment category sequence As an observation sequence, the purpose of the treatment is regarded as an implicit state, and the most likely sequence of hidden states is calculated according to the hidden Markov model; the abnormality output module is configured to be included in the implicit state included in the sequence of the implicit state In the case of an abnormal purpose, the output is related to the current and historical visit behavior of the risk control object Treatment data.

According to an embodiment of the present application, there is provided a computer readable storage medium having stored thereon a program for executing a risk control method for inferring a medical insurance violation by data analysis, the program, when executed by a processor, implements the following Step operation: Step 1. Obtain current and historical visit behavior data of the wind control object, and personal information and public data related to the wind control object; Step 2. Extract the visit with the risk control object from the data acquired in step 1. Behavior-related features; Step 3. According to the extracted features, each visit behavior in the current and historical visit behavior of the wind control object is divided into corresponding treatment categories to form a treatment category sequence; Step 4, will be in step 3. The formed medical treatment sequence is used as an observation sequence, and the purpose of the treatment is regarded as an implicit state, and the most likely implicit state sequence is calculated according to the hidden Markov model; step 5, if one of the most likely implicit state sequences is included Or a plurality of implied states corresponding to the abnormal purpose, outputting a doctor related to the current and historical visit behavior of the risk control object Data.

The beneficial effects of the present application are mainly as follows:

1. Improve the flexibility and adaptability of the medical insurance fund risk control;

2. Rehearsing possible violations through a combination of existing behaviors;

3. The relevant evidence chain is automatically retained in the process of risk identification and control for subsequent processing.

DRAWINGS

1 is a schematic flow chart of a risk control method for inferring a medical insurance violation behavior by data analysis according to an embodiment of the present application;

2 is a functional block diagram of a risk control system for inferring a medical insurance violation by data analysis according to an embodiment of the present application;

Figure 3 shows the overall flow of judging whether or not to include abnormal surveillance after calculating the implied sequence of treatment objectives based on the HMM model;

FIG. 4 illustrates an operating environment of a system in which an application is installed, according to an embodiment of the present application.

Detailed ways

The implementation of the technical solution will be further described in detail below with reference to the accompanying drawings.

It will be understood by those skilled in the art that the following description is to be construed as illustrative of the embodiments of the invention. The present application can be applied to other than the technical details exemplified below, as long as they do not depart from the principles and spirit of the present application.

In addition, in order to avoid limitation of the description of the present specification to the simplifications, in the description in the specification, some technical details that can be obtained in the prior art materials may be omitted, simplified, modified, etc., which is a technique in the art. It is understandable to the person and this does not affect the disclosure adequacy of this specification.

Hereinafter, embodiments of the present application will be described with reference to the drawings.

Note that the description will be given in the following order: 1. The risk control method for inferring medical insurance violations through data analysis (Figure 1); 2. The system for inferring medical insurance violations through data analysis (Figure 2-3); installed on a computer system application program for implementing an embodiment of the present application, and the application program storing medium readable described (FIG. 4).

1. Risk control method for inferring medical insurance violations through data analysis

1 is a flow chart showing a risk control method for inferring a medical insurance violation by data analysis according to an embodiment of the present application.

As shown in FIG. 1 , the wind control method for inferring medical insurance violation behavior through data analysis according to an embodiment of the present application mainly includes the following steps:

Step S100: Data acquisition, that is, acquiring various types of information related to data analysis, including insured (weather control object) information and public information.

The information of the insured person (the risk control object) includes the current and historical medical treatment behavior data of the insured person, and the personal information related to the insured person;

The insured person information includes medical insurance settlement data of the insured person, demographic information of the insured person's age, gender, culture, occupation, etc., and the public information includes the scale of the medical institution obtained by combining the public database, such as the company's industrial and commercial registration data, Geographical information, grade data, doctor's title, registration status, etc., the status of medical insurance fund audits and notifications from public information such as news, and case behaviors identified as problems in previous medical insurance fund audits, etc.

In step S200, the feature is extracted from the information acquired in the previous step, that is, the feature related to the insured person's visiting behavior is extracted from the data acquired in step S100.

Specifically, data features are extracted from the above various types of data for classification in the next step, and the extracted features are mainly classified into the following categories:

Demographic characteristics: including age, gender, occupation, and cultural level;

Geographical features: the insured person's place of origin, the company's geographic location, the location of the medical institution, the insured area, etc.;

Time characteristics: time of visit, interval between visits, time for admission;

Medical characteristics: diagnosis, treatment department, medical consumption list, medical institution scale, medical institution level, medical institution's previous case label, doctor's title, etc.;

Cost characteristics: single cost, total time period, cost component ratio (medical expenses, drug costs, inspection and inspection costs, material costs);

Step S300, establishing a classification of the medical treatment category, that is, according to the characteristics extracted in step S200, dividing each visiting behavior in the current and historical medical treatment behavior of the insured into the corresponding medical treatment category, forming a sequence of medical treatment categories,

Specifically, according to the feature extracted in step S200, the medical care settlement data is clustered into different types using the unsupervised learning method.

Step S400, establishing a hidden Markov model (HMM) model

The hidden Markov model is constructed with the insured as the axis.

Among them, the implied states constructed by the insured person's dimensions include: physical examination, initial diagnosis, follow-up, dispensing, hospitalization, and abnormality.

The medical treatment category established in step S300 is used as an observation sequence, and the hidden Markov model parameters (observation probability matrix and state transition matrix) are derived using the Baum-Welch algorithm.

Step S500: Using the HMM model established in step S400, output the inference result and the evidence chain according to the current behavior data of the insured and the historical behavior data.

Specifically, the treatment category sequence formed in step S300 is taken as an observation sequence, and the purpose of the treatment is taken as an implicit state, and the most likely implicit state sequence is calculated according to the hidden Markov model.

When, according to the observation probability matrix and the state transition matrix, it is inferred that an insured person has an abnormal medical treatment behavior, the corresponding insured person is output as a risk target, and the recorded behavioral state transition state is output as a evidence chain to the system audit. Or manual audit.

As an example, in step S100, the medical insurance settlement data and the demographic characteristics of the insured are provided by the social security settlement system of the implementation; the business registration information of the company can be obtained through the public channel of the website of the State Administration for Industry and Commerce of the People's Republic of China; the scale and geography of the medical institution The grade data can be obtained through the public websites of the local health and family planning committee website and the hospital homepage; the doctor's title and registration status can be obtained through the public channels of the local health and family planning committee websites; the audit status and notification status of the local medical insurance funds can be crawled through the news network. Collecting and collating; the case behavior identified as a problem in the previous medical insurance fund audit needs to be obtained from the data of the social security calendar year.

As an example, in step S200, the extracted features include the following categories.

Demographic characteristics: including age, gender, occupation, cultural level, work unit, etc., directly extracted from the social security database of the implementation, and the work unit is associated with the business registration information.

Geographical features: Through the network location provider, the distance between the insured person's place of origin, the company's geographic location, the geographic location of the medical institution, the participating area, and the geographical location are converted into coordinate values and values for storage. For example, the hometown "Shanghai" is input to the Baidu map API to obtain the GPS coordinates of Shanghai, and a certain community and a certain hospital are input map APIs to obtain the geographical distance between the two places.

Time characteristics: time of visit, interval between visits, time for admission. The time of visit and the time of insured are directly extracted from the data, and the interval between visits is the difference between the two visits.

Medical characteristics: diagnosis, treatment department, medical consumption list, medical institution scale, medical institution level, medical institution's previous case label, doctor title

Cost characteristics: single cost, total time period, cost component ratio (medical expenses, drug costs, inspection and inspection costs, material costs), etc.

As an example, in step S300, the features extracted in step S200 may be clustered into different clinic category categories using a K-means algorithm or an RVM classifier. For example, the treatment of cancer patients is divided into the initial diagnosis state based on the consumption test and the drug-based chemotherapy state.

As an example, in step S400, a hidden Markov model (HMM) model can be established as follows.

The different treatment categories were classified as the observation sequence O, and the purpose of the patient's visit (physical examination, initial diagnosis, referral, dispensing, hospitalization, abnormality) was used as an implicit state, and the Baum-Welch algorithm was used to solve the hidden Markov model parameters (observation probability). Matrix and state transition matrix).

Assuming that the medical treatment category is classified into four categories (visiting category) in step S300, the calculated observation state probability matrix and state transition probability matrix are respectively exemplified as follows (in a tabular manner for ease of understanding).

Table 1 observation state probability matrix

Table 2 state transition probability matrix

	Medical examination	New diagnosis	Review	Dispensing	Hospitalization	abnormal
Medical examination	0.32	0.01	0.13	0.34	0.19	0.01
New diagnosis	0.07	0.16	0.31	0.02	0.09	0.35
Review	0.08	0.37	0.07	0.26	0.20	0.02
Dispensing	0.07	0.25	0.03	0.29	0.05	0.31
Hospitalization	0.13	0.14	0.13	0.29	0.04	0.27
abnormal	0.07	0.34	0.27	0.15	0.12	0.05

The data in the above table (values in the matrix) is the probability obtained by statistical data, which can be summarized from known data obtained from medical institutions.

As an example, in step S500, based on the observation probability matrix and the state transition matrix calculated in step S400, the purpose of each visit of the patient can be dynamically inferred based on the patient visit behavior.

For example, if the observation sequence of a patient's previous 4 visits is O=(visit category 1, visit category 3, visit category 2, visit category 2), then according to the HMM model (not necessarily corresponding to the model above, it can be any Other HMM models) The most likely implied state of the reasoning (the purpose of the visit) can be I = (hospital, dispensing, initial diagnosis, initial diagnosis). Next, when the patient visits the 5th time, the visit category is “Visit Category 2”. At this time, the observation sequence becomes O=(visit category 1, visit category 3, visit category 2, visit category 2, visit category 2) According to the HMM model, the most likely sequence of implied states will be calculated as I= (hospitalization, dispensing, initial diagnosis, abnormality, initial diagnosis). Thus, when an abnormal state (corresponding to an abnormal medical purpose) occurs in the sequence of the implied state, the insured person can be included in the abnormal population (specially monitored population), and the hidden state sequence (sequence of the treatment purpose) and The observation sequence (study category sequence) is used as evidence output as evidence for system review or manual audit processing.

Optionally, the relevant insured person (the wind control object may also be according to the total number of occurrences of the abnormal state (corresponding to the abnormal purpose of the visit) in the sequence of the hidden state, and the calculated total probability of the hidden state sequence in which the abnormal state occurs. ) An abnormal population divided into different monitoring levels.

Optionally, the probability of occurrence of all possible implicit state sequences including the abnormal state may also be added to obtain a total probability including an abnormal state (an abnormal medical purpose), and if the total probability is higher than a predetermined threshold, The risk control object is assigned to a special monitoring group.

It should be noted that the data in the above table, and the specific content of the observation sequence O and the implicit state sequence I are only examples for illustrating the principle of the present application, and help those skilled in the art understand the manner of implementing the present application. It does not constitute a strict correspondence with the actual application, and the above data and specific content do not constitute any limitation on the present application.

As an alternative embodiment, as shown in FIG. 3, the suspected unreasonable medical treatment behavior (including the abnormal medical treatment purpose) may be divided into the following two cases:

1. The purpose of single or multiple visits is abnormal: as in the above example, the status marked as abnormal occurs. At this time, the risk control object is assigned to the special monitoring object by default, or system audit or manual audit processing is required;

2. Medical behavior sequence (hidden state sequence) anomaly: the calculated most likely implicit state sequence, although it does not contain an abnormal state, has an occurrence probability lower than a predetermined threshold. For example, the predetermined threshold may be set to a default of 25% of the probability of the most likely implied state sequence of equal length and can be changed as needed.

For example, the sequence of four consecutive medical treatments for a wind control object is (physical examination, physical examination, physical examination, physical examination). The probability of occurrence is lower than 25% of the most common sequence occurrence probability of 4 consecutive visits, and it is considered that the medical behavior sequence is abnormal. It is suggested that the behavior of the wind control object that occurs 4 times during the set time period for the purpose of physical examination is abnormal, and the characteristics related to the physical examination behavior of the risk control object (age, gender, medical treatment without medication, inspection) The inspection fee is high, the multiple inspection items are the same, the total cost is the same, etc.) is output as a chain of evidence to the background.

2. A system for inferring medical insurance violations through data analysis

According to an embodiment of the present application, there is provided a system for inferring a medical insurance violation behavior by data analysis, which is used to implement the above method, the system mainly comprising:

The wind control object data obtaining module is configured to acquire current and historical medical behavior data of the wind control object, and personal information and public data related to the wind control object;

a feature extraction module, configured to extract, from the data acquired by the wind control object data acquisition module, features related to a visiting behavior of the wind control object;

The medical treatment classification module is configured to divide each visit behavior in the current and historical visiting behaviors of the wind control object into corresponding medical treatment categories according to the extracted characteristics, and form a medical treatment category sequence;

a diagnosis target estimation module configured to use the treatment category sequence as an observation sequence, the treatment purpose as an implicit state, and the most likely implicit state sequence according to the hidden Markov model, including the wind control object The most likely purpose of the visit;

An abnormality output module configured to: when the most likely medical purpose of the wind control object included in the implicit state sequence corresponds to an abnormal purpose, the output is related to current and historical medical behavior of the wind control object Medical data.

According to an embodiment of the present application, the system may further include:

The hidden Markov model building module is configured to calculate the observation probability matrix and the state transition probability matrix corresponding to the hidden Markov model based on the big data acquired from the medical institution or the public data source by using the Baum-Welch algorithm.

3. A system for installing an application for implementing an embodiment of the present application, and storing the Computer readable medium for an application

Furthermore, various embodiments of the present application can also be implemented by a software module or computer readable instructions stored on one or more computer readable medium, where the computer readable instructions are executed by a processor or device component Different embodiments described herein are performed. Similarly, any combination of software modules, computer readable media, and hardware components are contemplated by the present application. The software modules can be stored on any type of computer readable storage medium such as RAM, EPROM, EEPROM, flash memory, registers, hard disk, CD-ROM, DVD, and the like.

In particular, another aspect of the present application relates to implementing the various embodiments described above using hardware and/or software. Those skilled in the art will appreciate that embodiments of the present application can be implemented or executed using a computing device or one or more processors. The computing device or processor can be, for example, a general purpose processor, a digital signal processor (DSP), an application specific integrated circuit (ASIC), a field programmable gate array (FPGA), or other programmable logic device, and the like. Different embodiments of the application may also be implemented or embodied by a combination of these devices.

Referring to Figure 4, there is shown an operating environment of a system in which an application is installed, in accordance with an embodiment of the present application.

In this embodiment, the system for installing the application is installed and runs in the electronic device. The electronic device may be a computing device such as a desktop computer, a notebook, a palmtop computer, or a server. The electronic device can include, but is not limited to, a memory, a processor, and a display. Figure 6 shows only the electronic device having the above components, but it should be understood that not all illustrated components may be implemented, and more or fewer components may be implemented instead.

The memory may be an internal storage unit of the electronic device, such as a hard disk or memory of the electronic device, in some embodiments. The memory may also be an external storage device of the electronic device in other embodiments, such as a plug-in hard disk equipped on the electronic device, a smart memory card (SMC), and a secure digital (Secure Digital) , SD) card, flash card (Flash Card), etc. Further, the memory may also include both an internal storage unit of the electronic device and an external storage device. The memory is used to store application software installed on the electronic device and various types of data, such as program code of the system in which the application is installed. The memory can also be used to temporarily store data that has been output or is about to be output.

The processor may, in some embodiments, be a central processing unit (CPU), a microprocessor, or other data processing chip for executing program code or processing data stored in the memory, such as performing the Install the application's system, etc.

The display may be an LED display, a liquid crystal display, a touch-sensitive liquid crystal display, an OLED (Organic Light-Emitting Diode) touch sensor, or the like in some embodiments. The display is for displaying information processed in the electronic device and a user interface for displaying visualizations, such as an application menu interface, an application icon interface, and the like. The components of the electronic device communicate with one another via a system bus.

Through the description of the above embodiments, those skilled in the art can clearly understand that the method in the foregoing embodiment can be implemented by means of software plus a necessary general hardware platform, and can also be implemented by hardware, but in many cases. The former is a better implementation. Based on such understanding, the technical solution of the present application in essence or the contribution to the prior art can be embodied in the form of a software product stored in a storage medium (such as ROM/RAM, magnetic The disc, the optical disc, includes a plurality of instructions for causing a terminal device (which may be a mobile phone, a computer, a server, an air conditioner, or a network device, etc.) to perform the method described in the various embodiments of the present application.

That is, according to an embodiment of the present application, there is also provided a computer readable storage medium storing thereon a program for executing a risk control method for inferring a medical insurance violation behavior by data analysis, the program being processed by a processor When executed, the steps according to the method are implemented.

It will be understood that the specific embodiments of the present invention have been described herein for the purpose of illustration, but the various modifications may be made without departing from the scope of the application. Those skilled in the art will appreciate that the operations and routines depicted in the flowchart steps or described herein can be varied in many ways. More specifically, the order of the steps may be rearranged, the steps may be performed in parallel, the steps may be omitted, other steps may be included, various combinations or omissions of routines may be made. Accordingly, the application is limited only by the accompanying claims.

Claims (20)

1. A wind control method for inferring medical insurance violation behavior through data analysis, characterized in that the purpose of the medical control object is divided into normal purpose and abnormal purpose, and the method comprises the following steps:

   Step 1. Obtain current and historical medical behavior data of the wind control object, and personal information and public data related to the risk control object;

   Step 2: extracting features related to the visiting behavior of the wind control object from the data acquired in step 1;

   Step 3: According to the extracted features, each visit behavior in the current and historical visiting behaviors of the wind control object is divided into corresponding medical treatment categories to form a medical treatment category sequence;

   Step 4: taking the sequence of the treatment category formed in step 3 as an observation sequence, using the purpose of the treatment as an implicit state, and calculating the most likely sequence of hidden states according to the hidden Markov model;

   Step 5: If the one or more implicit states included in the most likely implicit state sequence correspond to an abnormal purpose, or the probability of occurrence of the most likely implicit state sequence is lower than a predetermined threshold, then output and Medical data related to the current and historical visit behavior of the risk control object.
2. The wind control method for inferring medical insurance violation behavior by data analysis according to claim 1, wherein the hidden Markov model in step 4 comprises an observation probability matrix and a state transition probability matrix,

   The probability of each medical treatment category under each medical purpose is recorded in the observation probability matrix, and the probability of transferring from one medical treatment purpose to another is recorded in the state transition probability matrix.
3. The method of claim 1 for inferring a medical insurance violation by data analysis according to claim 1, wherein the step 5 further comprises:

   If the sum of the occurrence probabilities of all implied state sequences containing the anomalous purpose is above a predetermined threshold, the wind control object is identified as a risk target and output to the backend system.
4. The risk control method for inferring medical insurance violation behavior by data analysis according to claim 2, wherein in step 3, using K-Means or RVM classifier, the current and historical visiting behavior of the wind control object is Each visit is classified into the corresponding treatment category.

   Among them, the personal data related to the risk control object includes the medical insurance settlement data of the wind control object and the demographic data of the risk control object.

   Public data related to risk control objects include data related to medical institutions and medical insurance fund audit data.
5. The risk control method for inferring medical insurance violation behavior by data analysis according to claim 2, wherein the normal medical treatment purpose comprises: physical examination, initial diagnosis, referral, dispensing, hospitalization,

   The observation probability matrix and the state transition probability matrix are derived using the Baum-Welch algorithm.
6. The risk control method for inferring a medical insurance violation behavior by data analysis according to claim 1, wherein in the step 2, the feature comprises:

   Demographic characteristics, including the age, gender, occupation, and cultural level of the risk control object;

   Geographical characteristics, including the birthplace of the risk control object, the geographical location of the company, the geographical location of the medical institution, and the participating area;

   Time characteristics, including the time of visit of the risk control object, the interval between visits, and the time of participation;

   Medical characteristics, including diagnostic data of the risk control object, the visiting department, the medical consumption list, the size of the medical institution, the level of the medical institution, the previous case label of the medical institution, and the average title of the doctor;

   Cost characteristics, including the single cost of the risk control object, the total cost of the time period, and the proportion of the cost.
7. The risk control method for inferring medical insurance violation behavior by data analysis according to claim 1, wherein in step 5,

   If one or more of the medical purposes of the risk control object included in the most likely implicit state sequence correspond to an abnormal purpose, the risk control object is identified as a risk target and output to the backend system.
8. A risk control method for inferring a medical insurance violation behavior by data analysis according to claim 1, wherein in step 5,

   If the most likely implicit state sequence does not contain an abnormal purpose, but the probability of occurrence of the most likely implicit state sequence is below a predetermined threshold, the wind control object is identified as a risk target and output to the back end system.
9. A wind control system for performing a risk control method for inferring medical insurance violations through data analysis, comprising:

   The wind control object data obtaining module is configured to acquire current and historical medical behavior data of the wind control object, and personal information and public data related to the wind control object;

   a feature extraction module, configured to extract, from the data acquired by the wind control object data acquisition module, features related to a visiting behavior of the wind control object;

   The medical treatment classification module is configured to divide each visit behavior in the current and historical visiting behaviors of the wind control object into corresponding medical treatment categories according to the extracted characteristics, and form a medical treatment category sequence;

   a diagnosis target estimation module configured to use the treatment category sequence as an observation sequence, the treatment purpose as an implicit state, and calculate a most likely implicit state sequence according to the hidden Markov model;

   An abnormality output module configured to output medical data related to current and historical visit behavior of the risk control object if one of the implicit states included in the implicit state sequence corresponds to an abnormal purpose; as well as

   The hidden Markov model building module is configured to calculate the observation probability matrix and the state transition probability matrix corresponding to the hidden Markov model based on the big data acquired from the medical institution or the public data source by using the Baum-Welch algorithm.
10. The wind control system according to claim 9, wherein the hidden Markov model comprises an observation probability matrix and a state transition probability matrix,

    The probability of each medical treatment category under each medical purpose is recorded in the observation probability matrix, and the probability of transferring from one medical treatment purpose to another is recorded in the state transition probability matrix.
11. The wind control system of claim 9 wherein said features comprise:

    Demographic characteristics, including the age, gender, occupation, and cultural level of the risk control object;

    Geographical characteristics, including the birthplace of the risk control object, the geographical location of the company, the geographical location of the medical institution, and the participating area;

    Time characteristics, including the time of visit of the risk control object, the interval between visits, and the time of participation;

    Medical characteristics, including diagnostic data of the risk control object, the visiting department, the medical consumption list, the size of the medical institution, the level of the medical institution, the previous case label of the medical institution, and the average title of the doctor;

    Cost characteristics, including the single cost of the risk control object, the total cost of the time period, and the proportion of the cost.
12. The air control system according to claim 9, wherein the abnormality output module is further configured to: when the sum of occurrence probabilities of all implied state sequences including an abnormal purpose is higher than a predetermined threshold The wind control object is identified as a risk target and output to the backend system;

    Or configured to identify the risk control object as a risk target and output when one or more of the medical purposes of the risk control object included in the most likely implicit state sequence correspond to an abnormal purpose To the backend system;

    Or configured to identify the wind control object as a risk target and output when the most likely implied state sequence does not contain an abnormal purpose, but the probability of occurrence of the most likely implicit state sequence is below a predetermined threshold To the backend system.
13. A computer readable storage medium having stored thereon a program for executing a risk control method for inferring a medical insurance violation behavior by data analysis, the program being executed by a processor, implementing the operations of the following steps:

    Step 1. Obtain current and historical medical behavior data of the wind control object, and personal information and public data related to the risk control object;

    Step 2: extracting features related to the visiting behavior of the wind control object from the data acquired in step 1;

    Step 3: According to the extracted features, each visit behavior in the current and historical visiting behaviors of the wind control object is divided into corresponding medical treatment categories to form a medical treatment category sequence;

    Step 4: taking the sequence of the treatment category formed in step 3 as an observation sequence, using the purpose of the treatment as an implicit state, and calculating the most likely sequence of hidden states according to the hidden Markov model;

    Step 5: If the one or more implicit states included in the most likely implicit state sequence correspond to an abnormal purpose, or the probability of occurrence of the most likely implicit state sequence is lower than a predetermined threshold, then output and Medical data related to the current and historical visit behavior of the risk control object.
14. The computer readable storage medium according to claim 13, wherein the hidden Markov model in step 4 comprises an observation probability matrix and a state transition probability matrix,

    The probability of each medical treatment category under each medical purpose is recorded in the observation probability matrix, and the probability of transferring from one medical treatment purpose to another is recorded in the state transition probability matrix.
15. The computer readable storage medium of claim 13, wherein the step 5 further comprises:

    If the sum of the occurrence probabilities of all implied state sequences containing the anomalous purpose is above a predetermined threshold, the wind control object is identified as a risk target and output to the backend system.
16. The computer readable storage medium according to claim 14, wherein in step 3, each visit behavior in the current and historical visit behavior of the wind control object is divided into corresponding uses using a K-Means or RVM classifier Visit category,

    Among them, the personal data related to the risk control object includes the medical insurance settlement data of the wind control object and the demographic data of the risk control object.

    Public data related to risk control objects include data related to medical institutions and medical insurance fund audit data.
17. The computer readable storage medium according to claim 14, wherein the normal medical purpose includes: physical examination, initial diagnosis, follow-up, dispensing, hospitalization,

    The observation probability matrix and the state transition probability matrix are derived using the Baum-Welch algorithm.
18. The computer readable storage medium of claim 13, wherein in step 2, the feature comprises:

    Demographic characteristics, including the age, gender, occupation, and cultural level of the risk control object;

    Geographical characteristics, including the birthplace of the risk control object, the geographical location of the company, the geographical location of the medical institution, and the participating area;

    Time characteristics, including the time of visit of the risk control object, the interval between visits, and the time of participation;

    Medical characteristics, including diagnostic data of the risk control object, the visiting department, the medical consumption list, the size of the medical institution, the level of the medical institution, the previous case label of the medical institution, and the average title of the doctor;

    Cost characteristics, including the single cost of the risk control object, the total cost of the time period, and the proportion of the cost.
19. The computer readable storage medium of claim 13, wherein in step 5,

    If one or more of the medical purposes of the risk control object included in the most likely implicit state sequence correspond to an abnormal purpose, the risk control object is identified as a risk target and output to the backend system.
20. The computer readable storage medium of claim 13, wherein in step 5,

    If the most likely implicit state sequence does not contain an abnormal purpose, but the probability of occurrence of the most likely implicit state sequence is below a predetermined threshold, the wind control object is identified as a risk target and output to the back end system.

Images