CN101697167B - Clustering-decision tree based selection method of fine corn seeds - Google Patents

Abstract

The invention discloses a clustering-decision tree based selection method of fine corn seeds, comprising the following steps: preprocessing data; establishing a decision tree; systematically judging the class of an input attribute value according to the established decision tree, and obtaining a point with a minimum distance from a three-dimensional point by computing the distances between the three-dimensional point and other points in the class, wherein the attribute of the point is most approximated to the corn attribute of an input corn seed; and querying the parent class and the mother class of the input corn seed from a children table so as to complete a corn seed selection function. The invention can optimize the fine corn seeds in the generational information, the growth information, the harvest information, and the like of corns according to requirements by combining a clustering and decision tree algorithm and can achieve the purposes of reducing the labor intensity and enhancing the decision efficiency.

Description

Selection method of fine corn seeds based on clustering-decision tree

Technical field:

The present invention relates to a kind of selection method of fine corn seeds.

Background technology:

Cluster is a kind of common data analysis tool, is meant with the method research of mathematics and handles given object, and be a kind of of multivariate statistical analysis.Thought based on " things of a kind come together, people of a mind fall into the same group ", it the set divide into several classes of mass data point or bunch, make between the data in each class farthest similarly, and the data in the inhomogeneity are farthest different, thereby find the distribution pattern of the overall situation and the mutual relationship between the data attribute.An outstanding feature of cluster analysis is to handle huge, complicated data set, but also can be used as the pre-treatment step of other algorithms.Clustering algorithm at present commonly used also has that k-is average, k-mould, k-central point, DIANA, AGNES, STING, COBWEB etc.

These classical clustering algorithms can only be handled simple, general data, and are not good enough for large-scale, complicated data set effect, so a lot of improved and new algorithm constantly is suggested.Such as, have document proposed a kind of evolution based on polycentric dynamic clustering algorithm, wherein bunch number need not pre-establish, for different bunch optional majority central points, and assess by two objective functions, improved the application of algorithm for data set with special characteristics and special distribution; The CUZ algorithm has improved the calculating of representative point, can not only differentiate data non-circular, non-rectangular shape effectively, can also the non-convex shape of treatment surface bunch; The GriDBSCAN algorithm is by at data space structure grid, the local earlier density algorithm that adopts, back polymerization generation real bunch, improved the efficient of DBSCAN greatly, reduced complexity.Also have some new algorithms, cohesion hierarchical clustering algorithm such as people such as Yu Wei proposition, adopted the spatial sequences of 84 dimensions to come indicated object,, finally can obtain than the efficient system tree graph by the matrix that is made of the Pearson's coefficient between object is in twos condensed; The novel ant algorithm that on the basis of population algorithm and intelligent body, proposes, the number that needn't stipulate in advance bunch, and can be by calculating local object but not all the method for objects reduce the whole calculated amount of algorithm, can find arbitrary shape bunch.

Traditional decision-tree is to use one of logical method more widely in the sorting technique, it infers the classifying rules of decision tree representation from one group of out of order, random example, it adopts top-down recursive fashion, utilize the mutual information (information gain) in the information theory to seek the field that has maximum fault information in the database, set up the node of decision tree, again according to the branch of the different value inspections tree of field, in each descendent subset, repeat to set up lower floor's node and the ramifying of tree.Paths from the root of decision tree to leaf node just corresponding a conjunction rule, whole decision tree just corresponding one group of expression formula of extracting.Method commonly used has C4.5, ID3, CLS etc.

Proposed a lot of improved decision Tree algorithms at present, the interest parameter of having introduced the user when differentiating the degree of correlation of attribute at the computing information entropy such as people such as Ding Rongtao has been accelerated the speed that information entropy reduces to reduce the redundancy between attribute; People such as Fadila Bentayeb propose a kind of new data mining algorithm based on decision tree, by the pre-service of before handling in data mining data being carried out, set up a possibility table, and ID3 algorithm and the information gain formula new to this enforcement, simplify the complexity of algorithm, finally reached the purpose of decision tree classification; Concern sorting technique RDC algorithm, use the ID array to propagate and upgrade to concern decision tree more, can prevent to set up the relation table of excessive redundancy, prevented losing of information effectively, optimized tree construction, strengthened algorithm validity.

Infotech is widely used in China's agricultural production, and it has advanced the development of China's agricultural production greatly, becomes the important support strength that developing modern agriculture is produced.Aspect the seed selection of fine corn seeds, storing bulk information, as plant height, fringe height, the time of infertility, cell production etc.,, will produce guiding effect to Agricultural Development and reform if can therefrom excavate correct, reliable, useful rule.

Summary of the invention:

The object of the present invention is to provide and a kind ofly can be as requested carry out the preferred of fine corn seeds, can reach reduction labour intensity, improve the selection method of fine corn seeds based on clustering-decision tree of the efficiency of decision-making growth and development information, growth information and the results information etc. of corn.

Technical solution of the present invention is:

A kind of selection method of fine corn seeds based on clustering-decision tree is characterized in that: comprise the following steps:

(1) data pre-service: the three-dimensional point in the corresponding one-tenth of three attributes space that will select, utilize the k-average algorithm in the clustering algorithm, calculate measuring point all in children's table and the distance between the center of gravity, relatively big or small, the most all records are poly-to be two bunches, and make the bigger similarity that records in each bunch, and different bunches have bigger distinctiveness ratio;

(2) set up decision tree: the property value of selected three attributes of first discretize, divide them into three classes, and with mass of 1000 kernel as categorical attribute, divide classification and be labeled as low yield, middle product, high yield, after the input time of infertility, mass of 1000 kernel and these three attributes of cell production, according to the Euclidean distance formula:

$d(x_i,y_j)={\left(\underset{k=1}{\overset{m}{\mathrm{\Σ}}}{(x_{\mathrm{ik}}-x_{\mathrm{jk}})}^2\right)}^{\frac{1}{2}}$

Can judge which clustering cluster is this three-dimensional point belong to, after reaching a conclusion, this bunch carried out the data mining of ID3 algorithm, calculate attribute and the information gain value of cell production attribute in the time of infertility, and the attribute that will have bigger gain is as testing attribute, at the child node place of testing attribute segmentation, draw branch, divide whole record set, carry out successively again forming a decision tree of simplifying, wherein attribute S _kThe information gain formula be:

$G\left(S_K\right)=h_s\left(s_j\right)-\underset{k=1}{\overset{K}{\mathrm{\Σ}}}(\frac{n_k}{n_j}\×(-\underset{i=1}{\overset{c}{\mathrm{\Σ}}}\frac{n_{\mathrm{ik}}}{n_k}\×{\log }_2\left(\frac{n_{\mathrm{ik}}}{n_k}\right)))$

N wherein _jBe the occurrence number of total node, n _kBe that the prediction property value is V _kThe occurrence number of child node, n _IkBe that categorical attribute is C _iThe prediction property value be V _kThe occurrence number of child node;

(3) according to the decision tree that forms, system judges which classification the property value of being imported belongs to, and by calculating the distance of other points in this three-dimensional point and the affiliated classification, obtain a point that distance is minimum, the attribute that this point is had is exactly the corn attribute that approaches to import corn variety most, from children's table, inquire parent and female class of this corn variety, promptly finish the function of corn seed selection.

The present invention can carry out the preferred of fine corn seeds to growth and development information, growth information and the results information etc. of corn as requested in conjunction with cluster and decision Tree algorithms, can reach reduction labour intensity, improves the purpose of the efficiency of decision-making.

Selection method of fine corn seeds based on clustering-decision tree is according to the corn attribute specification of being imported, utilize the algorithm of cluster and decision tree in the data mining, find out the most similar close corn variety to this attribute, and then from children table, find out parent and female class of this this corn variety, and the essential information of parent and female class exported, thereby obtain the parent of corn attribute specification of input and the breeding of female class.

This selection method of fine corn seeds is based on the auxiliary realization of computer software, has reduced labour intensity in the artificial fine-variety breeding greatly, has improved the efficiency of decision-making and the accuracy of fine corn seeds seed selection.

The invention will be further described below in conjunction with drawings and Examples.

Fig. 1 is the final decision figure of embodiment.

Fig. 2 is clustering algorithm description figure.

Fig. 3 is decision Tree algorithms description figure.

Embodiment:

1. select sample set;

Original sample set comes from the selected kind experiment of certain Agricultural Information group 2006 and gathers (Y group) tables of data the end of the year.Because original sample set data message amount is big, corn variety is more, describes for convenience of description, now only selects these eight records of Y1-Y8 to discuss, and cited to choose sample set as shown in table 1 below.Table 1 is selected sample set

Number of times	Mean value (bunch 1)	Mean value (bunch 2)	Produce new bunch	New mean value (bunch 1)	New mean value (bunch 2)
						1	(100， 200.8， 6.73)	(101， 269.8， 7.83)	?{Y1，Y7}，{?Y2，Y3，Y4，?Y5，Y6，Y8}	(100， 213.9， 6.555)	(100.33， 275.23， 7.67)
2	(100， 213.9， 6.555)	(100.33， 275.23， 7.67)	?{Y1，Y6，Y7?}，{Y2，Y3，Y?4，Y5，Y8}	(100， 212.87， 6.75)	(100.4， 288.12，7 .782)
						3	(100， 213.9， 6.555)	(100.33， 275.23， 7.67)	?{Y1，Y6，Y7?}，{Y2，Y3，Y?4，Y5，Y8}	(100， 212.87， 6.75)	(100.4， 288.12，7 .782)

Table 2 cluster overall process

The corn variety code	Period (cycle)	Weight (weight)	Yield (output)
				Y1	100	200.8	6.73
Y2	101	269.8	7.83
				Y3	99	287.3	6.70
Y4	101	303.5	7.54
				Y5	100	255.5	7.56
Y6	100	210.8	7.13
				Y7	100	227.0	6.38
Y8	101	324.5	9.28

2. sample set is carried out cluster, as the pre-treatment step of decision Tree algorithms.Table 2 has provided mean value calculation and process of clustering and result in the whole process;

Iteration for the first time: two objects that supposition is selected at random, be used as initial point as Y1 and Y2, find respectively from 2 nearest objects, and produce two bunches of { Y1, Y7} and { Y2, Y3, Y4, Y5, Y6, Y8}.Bunch difference calculating mean value for producing obtains mean point.For { mean point is (100,213.9,6.555) for Y1, Y7}; For Y2, and Y3, Y4, Y5, Y6, Y8}, mean point is (100.33,275.23,7.67).

Iteration for the second time: adjust object place bunch by mean value, cluster is about to all and presses mean point (100,213.9,6.555) again, and (100.33,275.23,7.67) nearest principle is redistributed.Obtain two new bunch: { Y1, Y6, Y7} and { Y2, Y3, Y4, Y5, Y8}.Recomputate a bunch mean point, obtain new mean point and be (100,212.87,6.75) and (100.4,288.12,7.782).

Iteration for the third time: all are pressed from mean point (100,212.87,6.75) and (100.4,288.12,7.782) and nearest principle redistributes, and adjusts object, bunch still be { Y1, Y6, Y7} and { Y1, Y6, Y7}, find not occur redistributing, and the criterion function convergence, EOP (end of program).

Through three iteration, can obtain final two cluster C1, C2 class.Wherein, C1 comprises Y1, Y6, and Y7, C2 comprises Y2, Y3, Y4, Y5, Y8.

3. import alternative condition, classification under judging;

Import a group selection condition at random, suppose Period=101, Weight=290, Yield=7.2, then can correspond to the three-dimensional point P (101,290,7.2) in the space, calculate P o'clock distance according to the Euclidean distance formula to two bunches of centers, obtain d1=77.14, d2=2.06 (d1 is the distance that P arrives class C1 center of gravity, and d2 is the distance that P arrives class C2 center of gravity), because d1＞d2 can judge that will put P is included among the class C2.

4. the cluster of selecting is implemented the ID3 algorithm, select breeding; Determine that at first categorical attribute is Weight (sample of Weight＜=300 is a low yield, and the sample of Weight＞300 is a high yield), carries out segmentation (Period＜=100, Period＞100 to attribute Period and Yield then; Yield＜=7.6, Yield＞7.6).As shown in table 3.The attribute value table of table 3 sample

Y2	＞100	＞7.6	Low yield
				Y3	＜100	＜7.6	Low yield
Y4	＞100	＜7.6	High yield
				Y5	＜100	＜7.6	Low yield
Y8	＞100	＞7.6	High yield

The expectation information of calculating categorical attribute Weight is: $I\left(\mathrm{Weight}\right)=-\frac{3}{5}{\log }_2\left(\frac{3}{5}\right)-\frac{2}{5}{\log }_2\left(\frac{2}{5}\right)=0.97;$

The entropy of computation attribute Period again: $E\left(\mathrm{Period}\right)=0-\frac{3}{5}(\frac{1}{3}{\log }_2\left(\frac{1}{3}\right)+\frac{2}{3}{\log }_2\left(\frac{2}{3}\right))=0.55;$

The entropy of attribute Yield:

$E\left(\mathrm{Yield}\right)=-\frac{3}{5}(\frac{1}{3}{\log }_2\left(\frac{1}{3}\right)+\frac{2}{3}{\log }_2\left(\frac{2}{3}\right))-\frac{2}{5}(\frac{1}{2}{\log }_2\left(\frac{1}{2}\right)+\frac{1}{2}{\log }_2\left(\frac{1}{2}\right))=0.95;$

Then the information gain of attribute Period is: Gain (Period)=I (Weight)-E (Period)=0.42;

The information gain of attribute Yield is: Gain (Yield)=I (Weight)-E (Yield)=0.02;

Because attribute Period has bigger information gain, then with Period as testing attribute, create decision tree, obtain Fig. 1.

According to the alternative condition of importing at random before, can determine that this kind belongs to Y2 low yield class, only need parent and the female class of Y2 in the Query Database, can obtain parents' essential information of breeding expectation.

2 arthmetic statements

Selection method of fine corn seeds based on clustering-decision tree is summarized as follows: at first all samples are carried out cluster, obtain required bunch of class, the attribute of the object of judging as required is included in the corresponding bunch class then.Last at bunch class implementation decision tree algorithm among a small circle that finds, create best decision tree, thereby find separating of optimizing.Wherein the description of cluster and decision Tree algorithms is shown in Fig. 2 and 3.

According to selection method of fine corn seeds based on clustering-decision tree, utilization VC++ programming and ACCESS database technology, development and Design based on the software systems of the fine corn seeds seed selection of clustering-decision tree, comprised basic database, knowledge base and reasoning storehouse in the ACCESS database of system.Wherein, basic database comprises user message table and corn information table, and knowledge base has comprised parents' table and children's table, and the reasoning storehouse is to set up according to the rule that corn breeding expert breeding is for many years summarized the experience.

The corn information management module comprises two parts, the one, and information inquiry operation, the one, the interpolation of corn essential information, deletion, retouching operation.

The function that fine corn seeds seed selection module realizes is according to the corn attribute of being imported, utilize the algorithm of cluster and decision tree in the data mining, find out the corn variety the most similar with these attributes, and then from children table, find out parent and female class of this corn variety, and their essential information is outputed in the list box.

Described " corn information table (corn) " is used to write down the essential information of corn, comprises that the time of infertility, plant height, fringe height, mass of 1000 kernel, cell production, comparison attribute such as shine.

Table 4 corn information table (corn)

Field name	Data type	Could be sky	Constraint condition	Explanation
					Corn_id	Automatic numbering	Not	Major key	Numbering
Corn_name	Text	Not	\	The kind code
					Corn_period	Text	Not	\	The time of infertility
Corn_high	Text	Not	\	Plant height
					Corn_shigh	Text	Not	\	The fringe height
Corn_length	Text	Not	\	Spike length
					Corn_wide	Text	Not	\	Fringe is thick
Corn_weight	Text	Not	\	Mass of 1000 kernel
					Corn_yield	Text	Not	\	Cell production
Corn_percent	Text	Not	\	Comparison is shone

Described " knowledge base-parents show (father) " can show that parents indicate and child information etc., and be as shown in table 5.

Table 5 parents show (father)

Field name	Data type	Could be sky	Constraint condition	Explanation
					Pid	Automatic numbering	Not	Major key	Numbering
Pname	Text	Not	\	The kind code
					Pyield	Text	Not	\	Cell production
Pweight	Text	Not	\	Mass of 1000 kernel
					Gperiod	Text	Not	\	The time of infertility
Plogo	Text	Not	\	Parents' sign
					Pnum	Text	Not	\	Children's number
Pfirst	Text	Not	\	Initial children's code

" knowledge base-children show (son) " can show corresponding parent corn variety code, and be as shown in table 6.

Table 6 children show (son)

Field name	Data type	Could be sky	Constraint condition	Explanation
					Sid	Automatic numbering	Not	Major key	Numbering
Sname	Text	Not	\	The kind code
					Syield	Text	Not	\	Cell production
Sweight	Text	Not	\	Mass of 1000 kernel
					Speriod	Text	Not	\	The time of infertility
Fathername	Text	Not	\	Parent kind code
					Mothername	Text	Not	\	Female veriety code

Described " reasoning storehouse ", be according to corn breeding expert breeding for many years two rules of summarizing the experience out: the comparison of ◇ filial generation is according to＞=0; The breeding time of the breeding time＜parent of ◇ filial generation.From parents' table, select the parents of breeding corn.

Described " fine corn seeds seed selection module ", be according to the corn attribute specification of being imported, utilize the algorithm of cluster and decision tree in the data mining, find out the most similar close corn variety to this attribute, and then from children table, find out parent and female class of this this corn variety, and the essential information of parent and female class exported, thereby obtain the parent of corn attribute specification of input and the breeding of female class.This selection method of fine corn seeds is based on the auxiliary realization of computer software, has reduced labour intensity in the artificial fine-variety breeding greatly, has improved the efficiency of decision-making and the accuracy of fine corn seeds seed selection.

Claims (1)

1. the selection method of fine corn seeds based on clustering-decision tree is characterized in that: comprise the following steps:

(1) data pre-service: the three-dimensional point in the corresponding one-tenth of three attributes space that will select, utilize the k-average algorithm in the clustering algorithm, calculate measuring point all in children's table and the distance between the center of gravity, and it is relatively big or small, the most all records are poly-to be two bunches, and make the bigger similarity that records in each bunch, and different bunches have bigger distinctiveness ratio;

(2) set up decision tree: the property value of selected three attributes of first discretize, divide them into three classes, and with mass of 1000 kernel as categorical attribute, divide classification and be labeled as low yield, middle product, high yield, after the input time of infertility, mass of 1000 kernel and these three attributes of cell production, according to the Euclidean distance formula:

$d(x_i,y_j)={\left(\underset{k=1}{\overset{m}{\mathrm{\Σ}}}{(x_{\mathrm{ik}}-x_{\mathrm{jk}})}^2\right)}^{\frac{1}{2}}$

Can judge which clustering cluster is this three-dimensional point belong to, after reaching a conclusion, this bunch carried out the data mining of ID3 algorithm, calculate attribute and the information gain value of cell production attribute in the time of infertility, and the attribute that will have bigger gain is as testing attribute, at the child node place of testing attribute segmentation, draw branch, divide whole record set, carry out successively again forming a decision tree of simplifying, wherein attribute S _kThe information gain formula be:

$G\left(S_K\right)=h_s\left(s_j\right)-\underset{k=1}{\overset{K}{\mathrm{\Σ}}}(\frac{n_k}{n_j}\×(-\underset{i=1}{\overset{c}{\mathrm{\Σ}}}\frac{n_{\mathrm{ik}}}{n_k}\×{\log }_2\left(\frac{n_{\mathrm{ik}}}{n_k}\right)))$

N wherein _jBe the occurrence number of total node, n _kBe that the prediction property value is V _kThe occurrence number of child node, n _IkBe that categorical attribute is C _iThe prediction property value be V _kThe occurrence number of child node;

(3) according to the decision tree that forms, system judges which classification the property value of being imported belongs to, and by calculating the distance of other points in this three-dimensional point and the affiliated classification, obtain a point that distance is minimum, the attribute that this point is had is exactly the corn attribute that approaches to import corn variety most, from children's table, inquire parent and female class of this corn variety, promptly finish the function of corn seed selection.

Images