Research on data mining system based on artificial intelligence and improved genetic algorithm

3.1.1 An overview of association rule algorithm

For a long time, the link standard has become an important field of data extraction research, because it can find the relationship between attributes in the database, and it is one of the first problems to be considered in the extraction process. As shown in Table 1, the scope of application of connection rules is not only limited to sales data, but also increasingly uses connection rules to discover the relationship between various things and their meaning see Table 1 for details.

Where the support degree of the item set X={A,B} is the ratio of the number of transactions containing both the transaction A and the B to the total number of transactions. Use Sup (X) to represent the number of occurrences of itemsets X. That is, probability P (A∪B). Suppot ( A = > B ) = PA ∪ B ) = Sup ( X ) | D | = Sup ( A ∪ B ) | D |(1)

The ratio of the number of transactions containing A and B to the number of transactions containing A, that is, the confidence of the A=>B is {the degree of support of the A} divided by the degree of support of the item set {A,B}. That is, probability P (A |B). Confidence ( A = > B ) = P ( A | B ) = Sup ( A ∪ B ) Sup ( A )(2)

For the A=>B of this rule, the degree of interest is defined as:

- 20 ptInte ( A = > B ) = P ( A | B ) P ( A ) P ( B ) = Sup ( AB ) Sup ( A ) Sup ( B ) = Confidence ( A = > B ) Support ( B )(3)

Formula (3) when the interest is 1, it shows that the transaction A does not affect the transaction B, and the interest is more than 1, then the transaction A will lead to the transaction B, which is a rule. When interest is less than 1, the transaction A prevents the B. of the transaction

BP learning process of the pattern is as follows: provide training samples for the network pattern and disseminate data, that is, the difference between the actual output and the actual output. Through learning, the network is gradually stable and the network weight is changed one after another. The actual output of the network will tend to the ideal output and will be kept within the prescribed limit.

BP network is a network composed of access, output and intermediate networks, usually with only one default layer, in special cases, increasing the number of hidden layers as needed. The real data processing is as follows: the input vector of the input layer is usually the value of the sample information characteristic, and after the input layer receives the data, the data is transmitted directly to the ground, which is a set of data in the sample received. Because no special processing of data is required. Thus, the expression is as follows: X ⇀ = [ x 0 , x 1 , … , x n ](4)

Implicit layer: is an effective network processing layer, which usually adjusts the input layer by the activity function in a specific region, receives weighted data from the input layer, and processes the data through a threshold to obtain more data. An expression of the neural i of the hidden layer is as follows: x i = X ⇀ W i ⇀(5)

Connection weights between the hidden layer neuron i and the input layer are as follows: W i ⇀ = = [ w i 0 w i 1 w i 2 · . w in ](6)

And the expression of neuron i is: x i = [ x 0 w i 0 , x 1 w i 1 , … , x n w in ](7)

When the neuron i get the input data, it will be constrained to the expected value by the activation function. The activation function is S type activation function, as shown below. f ( x ) = 1 1 + e - ax ( 0 < f ( x ) < 1 )(8)

So the output value of the neuron i after the activation function F(·) constraint is: o i = f ( x i ) = f ( X ⇀ W ⇀ i )(9)

Hence the output vector of the whole hidden layer is:(where k is the number of neurons in the hidden layer) O ⇀ = [ o 1 , o 2 , … , o k ](10)

Output layer: the input of the output layer is the weighted sum of the output of the hidden layer, and the received data is processed by the activation function, and then the actual output of the whole network is obtained. An input expression j the output layer neuron is as follows: x j = O ⇀ W j ⇀(11)

Output layer is also a S activation function, which constrains the data of data output layer. and the output of the neuron j is: o j = f ( x j ) = f ( O ⇀ W j ⇀ )(12)

So the output vector of the whole network is: OO → = [ o 0 , o 1 , … , o m ] = F 2 ( F 1 ( X ⇀ W 1 ⇀ ) W 2 ⇀ )(13)

First, the mean square error function of the network is calculated, and its mathematical expression is as follows: E = 1 2 P ∑ P ∑ K ( t pk - o pk ) 2(14)

When the error of the network is obtained, the error first reaches the output layer and adjusts the weight of the output layer. The mathematical expression of weight matrix adjustment is as follows: Δ w ij = lr * δ * oh i w ij = w ij + Δ w ij δ j = oo j * ( 1 - oo j ) * ( y i - oo j )(15)

And after substitution, the above formula is simplified as follows: W jk ( t + 1 ) = W jk ( t ) + Δ W jk Δ W jk = - η ∂ E ∂ W jk δ k = ( t k - O k ) O k ( 1 - O k )(16)

The expected output of the general hidden layer is uncertain, and the adjustment formula of the weight matrix of the hidden layer is as follows: δ j = oh j * ( 1 - oh j ) * ∑ i = 1 m ( w ji * δ i o ) v ij = v ij + Δ v ij Δ v ij = lr * δ j * χ i(17)

The formula is simplified as follows: W ij ( t + 1 ) = W ij ( t ) + Δ W ij Δ W ij = - η ∂ E ∂ W ij = η δ j x i δ j = ∑ k δ k W jk O j ( 1 - O j ) O i(18)

Learn for the second time according to the next set of sample data and repeat until the network performance is optimal see Fig. 1 for details.

OPEN IN VIEWER

Fig.1

BP Network training process.

5.1.1 Coding

The coding of genetic algorithm is used to describe the feasible solution of the problem, that is, the feasible solution to decompose the spatial problem is transformed into the search space method which can be processed by genetic algorithm. In this paper, the coding method is adopted. The selection value of each service attribute is represented by the number after the decimal point, the number after each decimal point represents a gene, connects a service attribute, and forms a decimal string.

5.1.2 Design of fitness function

The degree of adaptation is usually used to measure the degree of excellence of a group to achieve or close to the optimization calculation. It is the basis of the application of genetic algorithm. The adaptive function is used to evaluate the degree of adaptation. The criteria for adaptive function are: F ( X ) = { aSupp ( X ) + β Conf ( X ) Inte ( X ) ⩾ 1 1 α Supp ( X ) + 1 β Conf ( X ) Inte ( X ) < 1(21)

The traditional roulette strategy is often immature. Therefore, this paper adopts the selection strategy based on immunization mechanism. The probability is as follows: P d = { 1 M ( 1 - d ) Individuals with the highest concentrations in the population 1 M ( 1 + d 2 1 - d ) Other individuals in groups(22)

The advantages of this approach are as follows: the higher the value of individual adaptation, the greater the possibility of PF adaptation, the greater the possibility of choosing P, the greater the possibility of individual selection (catalytic effect), and the higher the degree of convergence of the accelerated algorithm; the smaller the possibility of PF adaptation, the lower the degree of individual concentration, the lower the likelihood of commodity dependence and the lower the likelihood of commodity selection.

Cross operation is a process in which two matched chromosomes exchange a part of the gene in some way, resulting in two new chromosomes. individuals in this document, the intersection probability and the probability of polychlorinated triphenyl variability were dynamically adjusted using various methods. When the degree of adaptation is different, if the population difference is large, the heterogeneity is large, the possibility of crossover and mutation is small, and the population diversity is low, the adaptation tends to converge or optimize locally, it is allowed to change with the degree of adaptation. The possibility of interlacing and mutation increases, thus effectively preventing “premature” phenomena. The intersecting possibilities of self-adaptation used in this document are as follows: P c = { P c 1 f ′ < f avg P c 1 - P c 2 f ′ - f avg f max - f avg f ′ ⩾ f avg(23)

Variation is a simulation of gene mutations in biological evolution. In this paper, the following adaptive mutations are used: P m = { P m 1 - P m 2 f max - f f max - f avg f ′ ⩾ f avg P m 1 f ′ < f avg(24)

The basic idea of simulated annealing algorithm: from the point of view of statistical physics, with the decrease of temperature, the energy of matter will gradually approach a lower state, and finally reach a certain level, that is: P = { 1 f i + 1 ⩾ f i exp ( k ( f i + 1 - f i ) T ) f i + 1 < f i(25)

If the average adaptation of nearby generations is lower than a certain level, the flow algorithm of the above rules can improve the simulated annealing gene algorithm in Fig. 5: see Fig. 5 for details.

OPEN IN VIEWER

Fig.5

Flowchart of association rules of simulated annealing genetic algorithm.