A Novel Risk Warning Approach to Food Enterprises Using Data Mining and Information Fusion Technologies

The food enterprise risk warning problem is an important issue both in the theoretical and practical aspects. In order to improve the risk warning level of food enterprises, a novel risk warning approach is studied using some emerging technologies, such as, data mining and information fusion. For this reason, a novel risk warning technology is proposed to food enterprises based on data mining and information fusion. Experimental results suggest that this technology is feasible, correct and valid. Experimental results and trend analysis has many practical significance for assisting government management and decision of import and export food agricultural product safety.


INTRODUCTION
Data mining aims to analyze, extract rules and development trend among data, discover relations among discrete data, exclude interference of rubbish data and form useful knowledge.Some believe this is an advanced stage of Online Analytical Processing (OLAP).Data mining tool can utilize existing analysis tools to analyze relations among data in mass information and establish effective analysis models (Dai et al., 2014;Demetis and Angell, 2007).These models and relations can be used to carry out trend analysis.It is the process of seeking potential and useful information and knowledge among mass, noisy and irregular practical application data with incomplete information (Garcia, 2013;Pang et al., 2015).
In the face of mass import and export food and agricultural product inspection result information, entry and exit inspection and quarantine hopes to establish data warehouse of food monitoring information through integrating inspection results of testing organizations in each place and combine statistical analysis and data mining to achieve real-time monitoring and early warning of food safety state, evaluate food safety in an scientific and effective way, accurately predict food safety development trend and provide scientific decision basis for supervision institutions (Yager, 1987;Agus et al., 2010;Ross and Hann, 2007).For example, big data mining can discover regional and seasonal distribution rule information of a pesticide in food and calculate the development trend according to time change.
With the development of emerging technologies, some scientific approaches should developed to the risk warning of food enterprises.For this reason, a novel risk warning technology is proposed to food enterprises based on data mining and information fusion.

PROPOSED METHODS
In mass data gained from food and agricultural product safety inspection, much potential and useful information is hidden.Entry-exit inspection and quarantine department hopes to extract information for decision making, early warning and trend analysis through data mining, discover unqualified items of food and agricultural products and spatial distribution rules and predict development tendency.If pesticide application dosage in a region continues to rise and positive results increase year by year, the early warning information of export onions can be proposed.
To be more specific, multi-dimensional analysis of food name, type, production enterprise, region, inspection item, inspection result, time, disqualified items is carried out first.Main method involved in this study is to use SAS tool to mine association rules and establish data models so as to find the relevance or relation among item sets in mass inspection data.Association rules refer to inspection of correlative dependence among information.The discovery rules can be used to discover strong rules from data warehouse that Conk-dente ad support are the threshold value set in advance.SAS system is used to construct a complex but useful data structure-data cube according to data analysis demand.Association rule mining is applied to analyze effectiveness of inspection items, frequency and casual inspection, find casual inspection effectiveness of food and agricultural products and analyze regional distribution feature of positive result and time distribution rules.According to distribution characteristics discovered, spatial and temporal distribution rules, discovery rules are utilized to analyze the trend of safety risk factor.Trend prediction exceeding the preset value may serve as the early warning information.
Information fusion model adopted in this study is established on the basis of Back Propagation (BP) neural network, Support Vector Machine (SVM) and Dempster-Shafer (DS) evidence theory (it is called as IFCA).DS evidence theory is a method to handle uncertain information, which develops on the basis of Bayesian theory.It can handle uncertainty caused by both randomness and fuzziness.DS method can continuously narrow hypothesis set by depending on credibility distribution function accumulation and differentiate 'the unknown' and 'uncertainty'.Potential advantage of DS method is that it does not need prior probability and conditional probability density.In view of advantage in this aspect, it can be completely applied in uncertain assessment independently.DS evidence inference method is as follows.
Definition 1: We assume U is the set of propositions which consist of some mutually exclusive and exhaustive elements, called identification framework and means null proposition set.When the function m: 2 U → [0, 1] meets the following conditions: We call m(A) is basic trust measure of proposition A and m(A) means accurate trust degree of proposition A, i.e., direct explanation for A.
Definition 2: We assume U is an identification framework and m: 3) describes D-S composition rule.Since it meets people's several basic desire characteristics for composition rule (focus, commutativity and modularity), it is evidence composition rule which is most widely applied.Meanwhile, it is also the foundation of studying other composition rules.The principles of information fusion assessment model established by use of BP network, SVM and DS evidence theory are as follows.

EXPERIMENTAL RESULTS
To test accuracy and practicability of model, this study applies Matlab R2010a to process and classify real data under this model to gain final risk rating of users and known suspicious transactions so as to complete empirical test through comparison.

Information fusion:
A food enterprise in the current province is chosen.Corresponding attributes of customers are extracted and scoring results of subject risk index correspond to them.In combination of elementary probability assignment function of customer subject risk index, subject index risk comment set of a customer is gained, as shown in Table 1.
According to data preparation work in the initial period of experiment, these attribute values of all enterprises under different time series are concluded and imported.Matlab programming tool is used to calculate clustering outlier and then classification result and outlier degree of behavior indexes within a time series are gained.In combination of elementary probability assignment function of behavior indexes, behavioral index risk comment set of the enterprise is gained, as shown in Table 2.
According to the proposed approach, elementary probability assignment and weight of each index, risk fusion is conducted.The food enterprise risk assessment result are listed as shown in Table 3.It is known from Table 3 that main comment of food enterprise risk is low in recent 10 months and owns small uncertainty.

CONCLUSION
A novel risk warning technology is proposed to food enterprises based on data mining and information fusion.Experimental results suggest that this technology is feasible, correct and valid.

Definition 3 :
We assume Bel 1 and Bel 2 are two trust functions on identification framework U; m 1 and m 2 are the trust degree respectively; focal elements are

Table 1 :
Subject index risk comment set of food enterprise