CN105046362A - Real-time prediction method of food safety on the basis of association rule mining - Google Patents

Abstract

The invention discloses a real-time prediction method of food safety on the basis of association rule mining, and belongs to the field of food safety. The real-time prediction method comprises the following steps: firstly, using a mining method based on an association rule to find other food associated with problem food and relevancy between the problem food and other food; then, according to a food risk index system, measuring weight values and risk coefficients which affect the risk factors of the associated food; and finally, according to the weight values and the risk coefficients of the risk factors, calculating the risk values of the associated food. The method can find other food associated with the safety of certain problem food, can measure the frequencies of various risk factors which cause the certain food to generate a safety problem in a historical database to automatically regulate the weights of the risk factors which affect the food safety so as to calculate the comprehensive influence, i.e. a food safety degree, on the certain food by all risk factors, provides a food safety reference for consumers and provides a decision basis for decision makers.

Description

A real-time food safety prediction method based on association rule mining

technical field

The invention relates to the field of food safety, in particular to a real-time food safety prediction method based on association rule mining.

Background technique

The food safety traceability system was first established and perfected by the European Union in 1997 to deal with the "mad cow disease" problem. It emphasizes the unique identification and process tracking of products, and implements ISO9001 and other quality control methods for tracking and tracing in all aspects of product production, transportation, storage, and sales. Once a food safety problem occurs, the whereabouts of the food can be effectively traced. Recall unqualified products in time to minimize losses.

The existing food safety traceability system has a single function, and food traceability applications can only complete the collection of data related to food and its raw materials, as well as some intuitive and simple traceability and information statistics applications. For example, if a piece of pork sold in the market is found to be contaminated with microorganisms or to carry certain germs, the existing traceability system can trace where the piece of pork came from, but it cannot explain the cause of the microbial contamination or the emergence of germs, let alone predict the origin of the pork. Whether other pork in the same batch also has problems, and it is impossible to predict whether other foods processed with pork as raw materials also have safety problems. Microbial contamination or bacteria carried by pork may occur in farms, slaughtering workshops, transportation processes, sales links, etc. During this process, pork accidents in different links have different impacts on other foods. Existing systems cannot correlate other foods related to the problem food, let alone predict the safety level of other foods.

Association rule mining is an important direction in data mining technology, which was first proposed by AgrawalR. et al. It was originally designed to discover the relationship between data attributes from transaction databases. A typical application of it is market basket analysis. Association rule mining can dig out valuable knowledge describing the relationship between attributes in the data from a large amount of data.

Due to the time-series correlation, causal correlation and other relationships among food safety testing data, a data mining method based on association rule mining is adopted. The data mining method based on association rule mining can find other foods that meet the minimum support and minimum confidence related to the problem food.

Contents of the invention

In order to solve the problem that the food safety traceability system in the prior art cannot predict the safety of other related foods based on the safety of the problem food, the present invention provides a real-time food safety prediction method based on association rule mining, which can Carry out real-time monitoring, analysis and evaluation of food safety information in the system database, continuously update the degree of hazard of risk factors, and then predict the risk coefficient of food in real time according to the index system for measuring food risk, and display it intuitively in the form of graphics and numbers come out.

The technical solution adopted by the present invention to solve the above technical problems is: a real-time food safety prediction method based on association rule mining, comprising the following steps:

1) Collect safety information on the production and distribution of various foods, and enter this information into the system database;

2) Find out the food associated with the problem food through association rule mining;

3) Calculate the risk value of the relevant food;

4) Judge whether the food is safe according to the calculated risk value, and enter the information into the system database.

In the step 2), it is divided into three steps to find out the food associated with the problem food through association rule mining. The first step is to find out all frequent sets that meet the support degree. The second step is to use the frequent set to generate association rules. The third step The first step is to find out other foods related to the problem food that meet the minimum support and credibility through association rules; the specific operations are as follows:

Use the breadth-first algorithm Apriori to search the system database layer by layer, that is, use K-itemsets to explore (K+1)-itemsets, find out the set of frequent 1-itemsets, and record the set as , A collection for finding frequent 2-itemsets ,and used to find , and so on, until no frequent k-itemsets can be found:

set up It is a set composed of m different data items, where the elements are called items, and the collection of items is called an item set;

Given a transactional database , where each thing T is a subset of the itemset I, namely ;

is the total number of transactions in D, X and Y are both items or itemsets in T, ;

If the transaction T contains both X and Y, then the association rule can be obtained:

(1)

In the formula, The proportion of the transaction T that satisfies the condition in the transaction database D, that is, the support degree Support, is calculated as follows:

(2)

It is the proportion of transactions that include X in D and also include Y, that is, the confidence degree Confidence, and the calculation formula is as follows:

(3);

According to formula (2) and formula (3), other foods related to the problem food that meet the minimum support and credibility are calculated.

The specific operation of calculating the risk value of the relevant food in the step 3) is as follows:

First, the risk factors affecting food safety are divided into internal factors , external factors and additional factors , respectively calculate the internal factor , external factors and additional factors the weight of , and ,and ;

Among them, the internal factor the weight of for

(4)

external factors the weight of for

(5)

Second, the internal factors are calculated separately , external factors and additional factors risk factor , , ;

Among them, the internal factor risk factor for

(6)

In the formula, Indicates the first ingredients, Indicates the first the risk value of an ingredient, for the weight of , and ;

external factors risk factor for

(7)

in, , Respectively represent the risk value of the production, transportation, storage and sales environment of food, the calculation method of external factor risk value and weight is similar to the calculation method of food ingredient risk value and weight, adopt the method of weighted sum; in the present invention, the last Level risk value, that is, the risk value of harmful substances in the food risk index system, is evaluated by experts in the food field based on the degree of harm of food harmful substances, including the risk value represented by corporate reputation and consumer feedback. The risk value of the remaining stages is calculated from the risk value of its next level and the weight of the risk factor, and the weight of each stage The calculation formulas have been given in the examples.

additional factor risk factor for

(8)

In the formula, Indicates the credit risk value of food production enterprises, Indicates the risk value reflected by consumer feedback, , , , , The value of is set by the comprehensive evaluation of experts in the food field;

Finally, according to the formula Calculate the risk value of the food, where, , , represent the risk coefficients of internal factors, external factors, and additional factors, .

The idea of the present invention is: when a certain food has a safety problem, first, use the mining method based on association rules to find out other foods related to the problem food and the correlation between them, and then measure the influence of these foods according to the food risk index system. The weight value and risk coefficient of the risk factor of the relevant food, and finally calculate the risk value of the relevant food according to the weight value of the risk factor and the risk coefficient of the risk factor. If a risk indicator of a food is measured and found to be outside the normal range, the food will be directly classified as a high-risk food. Throughout the process, the weights of risk factors and risk coefficients are re-measured before use, and the latest values are updated into the database.

Since there are many factors affecting food safety, there are also many food risk indicators, and most of these indicators can be quantified into a specific value. Some of the indicators that affect food safety are harmful substances in food ingredients, and some are harmful substances in the environment that food comes into contact with during production, transportation, storage, and sales. In addition, consumer feedback and the reputation of manufacturers can be obtained from The side reflects the safety degree of food. Here, all factors that affect food safety are collectively referred to as harmful substances.

Beneficial effects: the present invention proposes a real-time food safety prediction method based on association rule mining, which can find other foods related to a certain food safety problem, and can measure various food safety problems in the historical database. The number of risk factors automatically adjusts the weight of these risk factors that affect food safety, so as to calculate the comprehensive impact of all risk factors on a certain food, that is, the degree of food safety, providing food safety reference for consumers and decision-making basis for decision makers wait. At the same time, the present invention can also monitor, analyze and evaluate the food safety information in the system database in real time, continuously update the degree of hazard of risk factors, and then predict the risk coefficient of food in real time according to the index system for measuring food risk, and use graphics and The numbers are displayed intuitively, so that when a certain food has a problem, whether there is a safety hazard in the related food.

Description of drawings

Fig. 1 is the food risk index system diagram in the embodiment of the present invention;

Fig. 2 is the calculation process of food risk value in the embodiment;

Fig. 3 is the curve chart of food risk prediction in the embodiment.

Detailed ways

A real-time food safety prediction method based on association rule mining, comprising the following steps:

1) Collect safety information on the production and distribution of various foods, and enter this information into the system database;

2) Find out the food associated with the problem food through association rule mining;

Finding the food associated with the problem food through association rule mining is divided into three steps. The first step is to find out all frequent sets that meet the support degree. The second step is to use the frequent set to generate association rules. The third step is to find out through association rules. Other foods related to the problematic food that meet the minimum support and credibility; the specific operations are as follows:

Use the breadth-first algorithm Apriori to search the system database layer by layer, that is, use K-itemsets to explore (K+1)-itemsets, find out the set of frequent 1-itemsets, and record the set as , A collection for finding frequent 2-itemsets ,and used to find , and so on, until no frequent k-itemsets can be found:

set up It is a set composed of m different data items, where the elements are called items, and the collection of items is called an item set;

Given a transactional database , where each thing T is a subset of the itemset I, namely ;

is the total number of transactions in D, X and Y are both items or itemsets in T, ;

If the transaction T contains both X and Y, then the association rule can be obtained:

(1)

In the formula, The proportion of the transaction T that satisfies the condition in the transaction database D, that is, the support degree Support, is calculated as follows:

(2)

It is the proportion of transactions that include X in D and also include Y, that is, the confidence degree Confidence, and the calculation formula is as follows:

(3);

According to formula (2) and formula (3), calculate other foods related to the problem food that meet the minimum support and credibility;

3) Calculate the risk value of the relevant food, the specific operation is as follows:

First, the risk factors affecting food safety are divided into internal factors , external factors and additional factors , respectively calculate the internal factor , external factors and additional factors the weight of , and ,and ;

Among them, the internal factor the weight of for

(4)

external factors the weight of for

(5)

Second, the internal factors are calculated separately , external factors and additional factors risk factor , , ;

Among them, the internal factor risk factor for

(6)

In the formula, Indicates the first ingredients, Indicates the first the risk value of an ingredient, for the weight of , and ;

external factors risk factor for

(7)

in, , Respectively represent the risk value of food production, transportation, storage, and sales environment. The calculation method of the external factor risk value and weight is the same as that of the internal factor risk value and weight, and both are calculated by weighted sum method;

additional factor risk factor for

(8)

In the formula, Indicates the credit risk value of food production enterprises, Indicates the risk value reflected by consumer feedback, , , , , The value of is set by the comprehensive evaluation of experts in the food field;

Finally, according to the formula Calculate the risk value of the food, where, , , represent the risk coefficients of internal factors, external factors, and additional factors, ;

4) Judge whether the food is safe according to the calculated risk value, and enter the information into the system database.

The above is the basic factual mode of the present invention, and the present invention will be further elaborated below in conjunction with specific embodiments.

When there is a food safety problem, first, use the mining method based on association rules to find out other foods related to the problem food and the correlation between them, and then measure the risk factors affecting these related foods according to the food risk index system. Finally, calculate the risk value of the relevant food according to the weight value of the risk factor and the risk coefficient of the risk factor. If a risk indicator of a food is measured and found to be outside the normal range, the food will be directly classified as a high-risk food. Throughout the process, the weights of risk factors and risk coefficients are re-measured before use, and the latest values are updated into the database.

1 Association rule mining

There are many mining algorithms for association rule analysis methods. According to the characteristics of uneven distribution of food testing item data, the breadth-first algorithm Apriori is selected. Its basic idea is: all non-empty subsets of frequent itemsets must also be frequent. Apriori uses an iterative method called layer-by-layer search, and K-itemsets are used to explore (K+1)-itemsets. First, find out the set of frequent 1-itemsets, which is denoted as . A collection for finding frequent 2-itemsets ,and used to find , and so on, until no frequent k-itemsets can be found. find each A database scan is required.

set up It is a collection of m different data items, where the elements are called items, and the collection of items is called an itemset. Given a transactional database , where each thing T is a subset of the itemset I, namely ; is the total number of transactions in D. Both X and Y are items or itemsets in T, . If the transaction T contains both X and Y, then the association rule can be obtained:

(1)

In the formula The calculation formula is as follows:

(2)

It is the proportion of transactions that contain X in D and also contain Y, that is, the confidence (Confidence), the calculation formula is as follows:

(3)

The goal of data mining is to select strong association rules that are greater than both the confidence threshold and the support threshold. Data mining is divided into two steps. The first step is to find out all frequent sets that meet the support degree; the second step is to use frequent sets to generate association rules. Since the first step needs to scan the transaction database multiple times, the consumption of time and space is the key to restrict the mining efficiency.

Use the association rule mining method to find out other foods that satisfy a certain relationship with the problem food. For example, if there are pathogens in pork on the market, you can find out ham sausage, sauced trotters and other foods that use pork as raw materials, and their correlation between. If the bagged sauced chicken wings sold in the supermarket are contaminated by bacteria, the related sauced chicken legs, marinated eggs, sauced duck necks, etc. may also have problems, because they contain many of the same ingredients, such as food flavors, salt, pickled peppers , food additives, etc. Think of food as a transaction, food ingredients as items of the transaction, through the transaction and affairs Perform mining and calculate correlation. In a word, other foods related to the problem food can be found through the association rule method and satisfy the minimum support and credibility.

2 Setting of risk value

In the present invention, the scope of food risk value is defined as ,in set to low risk, set to medium risk, Set to high risk, the range corresponding to the risk level is set by experts in the food field and can be adjusted. In the following cases, the risk level of the food is directly judged as high risk and the risk value is 10, and there is no need to carry out weighted calculation according to the food risk index system:

① Some harmful substances in food are restricted, and some are prohibited. As long as the restricted use is harmless to humans within the specified range, prohibited substances are not allowed. In the food sampling inspection, if there are prohibited harmful substances, the food risk value is directly set to 10;

② If a restricted hazardous substance in food exceeds a certain value, the food risk value is directly set to 10;

③If the ratio of the number of complaints against a certain food to the total number of complaints of the category it belongs to in the past month exceeds , then the food risk value is directly set to 10;

④ If the unqualified rate of harmful substances in a food exceeds a certain value, the risk value of the food is directly set to 10, and the unqualified rate = (number of unqualified harmful substances/total number of inspections).

The above indicators are not fixed, and experts in the field can adjust them according to the actual situation, and some indicators can also be added. Correspondingly, some special circumstances can be set up to directly judge the food as medium risk.

3 risk value assessment

The weight of risk factors affecting food safety is obtained by statistically recording the historical records of food safety incidents. The weight indicates the degree of influence of a certain risk factor on food. The greater the weight, the more obvious the impact of the factor on food safety. For example, for the bagged sauced chicken legs sold in supermarkets, there are many risk factors affecting the safety of sauced chicken drumsticks. According to the food risk index system, the risk factors affecting sauced chicken legs are divided into internal factors , external factors and additional factors , the internal factors include: food flavors, salt, pickled peppers, food additives and other ingredients, and the external factors include: the pollution index of the raw chicken growth environment, the colony index of the sauce chicken leg production workshop, the temperature of the production workshop, and the links of transportation, storage and sales The air pollution index and ambient temperature that sauced chicken legs are exposed to, and additional factors include: the reputation of the production company, the number of consumer complaints, etc. Risk value of chicken drumsticks in sauce ,in , , represent the risk coefficients of internal factors, external factors, and additional factors, , has a weight of , has a weight of , has a weight of (constant, value evaluated from historical experience), and Respectively represent the number of safety incidents of chicken drumsticks in sauce caused by internal factors and external factors in the historical records, represents the sum of the two.

Among the internal factors, there are many sub-factors of internal factors, such as the above-mentioned example of sauced chicken legs, food flavor ,salt , pickled pepper , food additives , and the weights of these subfactors are: , , , ,in . Similarly, there are many sub-factors in the external factors, and the weight representation of these sub-factors is similar to that of the internal sub-factors.

In summary, the general expression formula of the internal factor weight is:

(4)

The general expression formula of external factor weight is:

(5)

The weight of the additional factor is , is a constant greater than 0 and less than 1. Its value is estimated by domain experts based on historical experience.

Value at Risk of Food Ingredients ,in Indicates that the food's ingredients; Indicates the No. The risk value of a harmful substance, which is set by domain experts; for the first the weight of a harmful substance, . The risk value of food affected by ingredients, that is, the risk value of internal factors:

(6)

in, Indicates the first ingredients, Indicates the first the risk value of an ingredient, for the weight of , and .

The risk value of the external factor is:

(7)

in, , Represents the risk value of food production, transportation, storage, and sales environments, the calculation method of risk value and weight of external factors is similar to the calculation method of risk value and weight of food ingredients, and their calculation methods are similar to the calculation method of risk value of food ingredients , using a weighted sum method.

The risk value for the additional factor is:

(8)

in, Indicates the credit risk value of food production enterprises, Indicates the risk value reflected by consumer feedback, , , , , The value of is set by the comprehensive evaluation of experts in the food field.

The risk value for a food is:

(9)

in, , , represent the risk coefficients of internal factors, external factors, and additional factors, .

4Real-time prediction of food risk value

Because the risk value of food is affected by many factors, it is changing every moment. The system's prediction of food safety is real-time, and the risk value of all foods is measured at regular intervals according to the set frequency, and the calculation results are recorded in the database for easy query. The system can set an early warning threshold for each food, and the system will automatically alarm when the risk value of the food exceeds the threshold. The food risk value prediction map is shown in Figure 2. The forecast results can also be displayed in graphics and numbers, and the forecast results and food safety trends can be seen at a glance. You can check the risk value and safety trend of food in the past week, month, year, etc. by selecting the viewing conditions, and you can also view the risk value and safety trend of a certain type of food or a certain type of food. The food risk prediction curve is shown in Figure 3.

Claims (3)

1. A food safety real-time prediction method based on association rule mining, is characterized in that, comprises the following steps: 1) Collect safety information on the production and distribution of various foods, and enter this information into the system database; 2) Find out the food associated with the problem food through association rule mining; 3) Calculate the risk value of the relevant food; 4) Judge whether the food is safe according to the calculated risk value, and enter the information into the system database. 2. A real-time food safety prediction method based on association rule mining according to claim 1, characterized in that, in the step 2), it is divided into three steps to find out the food associated with the problem food through association rule mining , the first step is to find out all the frequent sets that meet the support degree, the second step is to use the frequent set to generate association rules, and the third step is to find out other foods related to the problem food that meet the minimum support degree and credibility through the association rules; specifically The operation is as follows: Use the breadth-first algorithm Apriori to search the system database layer by layer, that is, use K-itemsets to explore (K+1)-itemsets, find out the set of frequent 1-itemsets, and record the set as , A collection for finding frequent 2-itemsets ,and used to find , and so on, until no frequent k-itemsets can be found: set up It is a set composed of m different data items, where the elements are called items, and the collection of items is called an item set; Given a transactional database , where each thing T is a subset of the itemset I, namely ; is the total number of transactions in D, X and Y are both items or itemsets in T, ; If the transaction T contains both X and Y, then the association rule can be obtained: (1) In the formula, The proportion of the transaction T that satisfies the condition in the transaction database D, that is, the support degree Support, is calculated as follows: (2) It is the proportion of transactions that include X in D and also include Y, that is, the confidence degree Confidence, and the calculation formula is as follows: (3); According to formula (2) and formula (3), other foods related to the problem food that meet the minimum support and credibility are calculated. 3. A real-time food safety prediction method based on association rule mining according to claim 1, characterized in that the specific operation of calculating the risk value of related foods in the step 3) is as follows: First, the risk factors affecting food safety are divided into internal factors , external factors and additional factors , respectively calculate the internal factor , external factors and additional factors the weight of , and ,and ; Among them, the internal factor the weight of for (4) external factors the weight of for (5) Second, the internal factors are calculated separately , external factors and additional factors risk factor , , ; Among them, the internal factor risk factor for (6) In the formula, Indicates the first ingredients, Indicates the first the risk value of an ingredient, for the weight of , and ; external factors risk factor for (7) in, , Indicate the risk value of food production, transportation, storage and sales environment respectively, the calculation method of risk value and weight of external factors is the same as the calculation method of risk value and weight of internal factors, both of which are calculated by weighted sum method; additional factor risk factor for (8) In the formula, Indicates the credit risk value of food production enterprises, Indicates the risk value reflected by consumer feedback, , , , , The value of is set by the comprehensive evaluation of experts in the food field; Finally, according to the formula Calculate the risk value of the food, where, , , represent the risk coefficients of internal factors, external factors, and additional factors, .