JPH07244668A - Data analysis system - Google Patents

Abstract

(57) [Summary] [Structure] Feature storing and compressing means (1) for compressing a database while retaining features, and three-dimensional plot displaying means (2) for displaying records as points on an arbitrary three-dimensional space.
A three-dimensional shape display means (3) for displaying a predetermined three-dimensional shape determined by two or more specific points on the space on the space; and a record for detecting a record included in the three-dimensional shape. A data analysis system comprising a detection means (4) and a distinction display means (5) for distinguishing and displaying the detected records. [Effect] From a database consisting of a large number of records,
It is possible to generate a partial database of a discriminable scale while preserving characteristics such as the distribution status of records, and to efficiently extract characteristic record groups while grasping the tendency of all records.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention In the field of physical experiments or business, the present invention derives a relation between attributes of records contained in a database consisting of a large number of records, or uses a relation derived from the relation. The present invention relates to a data analysis system that predicts an output with respect to an input of.

[0002]

2. Description of the Related Art As the first conventional technology according to the present invention,
"Visual Data Statistical Analysis" (SAS) (SAS Institute Japan SAS System Brochure MJ
UX6P1 92NOV), when a record group is three-dimensionally plotted and a characteristic record group is extracted, it can be realized by operating a predetermined plane figure on the screen with a mouse.

As a second conventional technique according to the present invention, "L
BG algorithm "(Y. Linde, A. Buzo and RMGray:
"An Algorithm for Vector Quantization", IEEE Tran
s.Commun., COM-28,1, pp.84-95 (Jan.1980)), when compressing a database while preserving features, consider records as multidimensional vectors. , The multidimensional space is divided into predetermined regions, the initial value of the quantized representative vector for each region is determined, and the LBG algorithm is used for convergence.

As a third conventional technique according to the present invention, when a three-dimensional figure is input as described in "Three-dimensional coordinate input device" (Japanese Patent Laid-Open No. 63-124173),
Select one from the XY, YZ, and ZX coordinate planes, input the two-dimensional projection diagram of the three-dimensional figure onto the selected coordinate plane, and then, in the other coordinate planes, also obtain the two-dimensional projection diagram of each plane. It can be realized by inputting.

[0005]

In the first prior art described above, when a large number of records are displayed in a three-dimensional plot and a record group of a characteristic region is extracted, the displayed plots overlap each other. However, there is a problem that the distribution of records cannot be grasped sufficiently and it is difficult to extract a desired record group.

In the second prior art, when compressing a large number of records, it takes a very long time to process depending on the convergence condition, and there is a possibility that exceptional records may not be extracted. there were.

Further, in the third prior art, when a three-dimensional figure for extracting a characteristic record from a record group displayed in a three-dimensional plot is designated, X
Select one from the Y, YZ, and ZX coordinate planes, enter the two-dimensional projection drawing of the three-dimensional figure onto the selected coordinate plane, and
There is a problem that it is difficult to specify a figure including a desired record group because it is necessary to input a two-dimensional projection diagram on each of the other coordinate planes.

A first object of the present invention is to provide a data analysis system which can efficiently extract a characteristic record group from a database consisting of a large number of records. A second object of the present invention is to provide a data analysis system capable of creating a partial database that reflects the characteristics such as the distribution status of records as faithfully as possible from a database consisting of a large number of records.

A third object of the present invention is to provide a data analysis system capable of extracting characteristic records from a record group displayed in a three-dimensional plot.
A fourth object of the present invention is to provide a data analysis system capable of extracting and storing characteristic record groups from a database consisting of a large number of records.

A fifth object of the present invention is to provide a data analysis system capable of removing exceptional record groups from a database consisting of a large number of records.

[0011]

The above object can be achieved by the following constitution. In the first preferred embodiment of the present invention, the data analysis system includes (1) feature storage / compression means, (2) three-dimensional plot display means, (3) three-dimensional shape display means, (4) record detection means, ( 5) Consists of a distinction display means. The data analysis system having this configuration is referred to as a first data analysis system for convenience.

In a second preferred embodiment of the present invention, the feature storing and compressing means of the first data analysis system comprises:
A record is randomly extracted from the database, the distance between this record and other records is calculated, and the record whose distance is smaller than a predetermined threshold is represented by the extracted record. Remove the records from the record extraction candidates and add the extracted records to the partial database. The data analysis system having this configuration is referred to as a second data analysis system for convenience.

In the third preferred embodiment of the present invention, the three-dimensional shape display means of the first data analysis system defines a three-dimensional area with two specific two points at both ends of the diagonal line and each side of X and Y. , A cuboid parallel to any of the Z axes, and a sphere having a specific one point as the center and the other one as a point on the spherical surface are displayed. The data analysis system having this configuration is referred to as a third data analysis system for convenience.

In a preferred fourth embodiment of the present invention, the record detection means of the first data analysis system is
The records included in the three-dimensional shape are saved as a new partial database. The data analysis system having this configuration is referred to as a fourth data analysis system for convenience.

In a fifth preferred embodiment of the present invention, the record detection means of the first data analysis system is
The record included in the three-dimensional shape is deleted from the database and saved as a new partial database.
The data analysis system having this configuration is referred to as a fifth data analysis system for convenience.

[0016]

In the first data analysis system of the present invention, the feature storing / compressing means can generate a partial database compressed from a large amount of databases while retaining the features. Then, the record group in the partial database can be displayed as a point in an arbitrary three-dimensional space by the three-dimensional plot display means. Then, the three-dimensional shape display means can display a predetermined three-dimensional shape determined by two or more points on the three-dimensional space specified by the user on the three-dimensional space. Then, the record detection means can detect the record included in the three-dimensional shape. Then, the distinguishing display means can distinguish and display the detected records. Therefore, the first object of the present invention is to provide a data analysis system capable of efficiently extracting a characteristic record group from a database composed of a large number of records.

In the second data analysis system of the present invention,
The feature storing / compressing means randomly extracts the records, calculates the distance between the extracted record and the other records, and the record whose distance is smaller than a predetermined threshold is the extracted record. It is represented as a representative, removed from the subsequent record extraction candidates, and the extracted record is added to the partial database. Therefore, from the database of a large number of records, which is the second object of the present invention,
It is possible to provide a data analysis system capable of creating a partial database that reflects the characteristics such as the distribution status of records as faithfully as possible.

In the third data analysis system of the present invention,
The three-dimensional shape display means, as a three-dimensional area, is a rectangular parallelepiped in which two points specified by the user are points at both ends of a diagonal line and each side is parallel to any of the X, Y, and Z axes, or the user specifies. It is possible to display a sphere with one point as the center and another point as a point on the sphere. Therefore, the third object of the present invention is to provide a data analysis system capable of extracting characteristic records from a record group displayed in a three-dimensional plot.

In the fourth data analysis system of the present invention,
The record detection means detects a record included in the three-dimensional shape and saves it as a new partial database. Therefore, a fourth object of the present invention is to provide a data analysis system capable of extracting and storing characteristic record groups from a database consisting of a large number of records.

In the fifth data analysis system of the present invention,
The record detection means detects a record included in the three-dimensional shape, deletes it from the database, and saves it as a new partial database. Therefore, a fifth object of the present invention is to provide a data analysis system capable of removing exceptional record groups from a database consisting of a large number of records.

[0021]

Embodiments of the present invention will now be described in detail with reference to the drawings.

FIG. 1 is a conceptual diagram of a data analysis system showing a first embodiment of the present invention.
It includes three-dimensional plot display processing 2, three-dimensional shape display processing 3, record detection processing 4, and distinctive display processing 5.

In the feature preservation / compression process 1, a large number of databases are used to generate a partial database which is compressed while retaining its features. 3D plot display processing 2
In, the records in the partial database are displayed on the screen as points in the arbitrary three-dimensional space designated by the user. In the three-dimensional shape display process 3, the user designates two or more points three-dimensionally displayed on the screen with an input device such as a mouse, and a predetermined three-dimensional shape specified by these points is displayed. In the record detection processing 4, a record represented by a point included inside the three-dimensional shape is detected. In the distinctive display processing 5, the points representing the detected record are displayed separately from other points. The series of processes described above can be repeatedly performed, and the process can be easily returned to the previous process.

FIG. 2 is a hardware configuration diagram of the data analysis system of FIG. 1, which includes an input device 21, a central processing unit 22,
It comprises a display device 23 and an external storage device 24.

In FIG. 2, a record is read from the database in the external storage device 24 into the central processing unit 22, and in the central processing unit 22, the above-mentioned feature preservation compression processing 1, three-dimensional plot display processing 2, three-dimensional shape display. Processing 3, record detection processing 4, and distinctive display processing 5 are performed, and the user specifies the three-dimensional display axis from the input device 21.
The points for specifying the three-dimensional shape are designated, and the display device 2
In 3, a three-dimensional plot of the record, a three-dimensional shape, etc. are displayed.

Table 1 is an example of a database. A database of bank customer information, with attributes such as customer name, age, term balance, ordinary balance, reserve balance, number of term accounts, ...
・ Transaction years, number of transactions, periodicity balance changes. The lower wavy line portion 11 has a customer code of “o31”, an age of 41, a time deposit balance of 4010K ¥, an ordinary deposit balance of 2660K ¥, a savings deposit balance of 6975K ¥, ...
It shows that the number of transactions is 133 and the change in periodicity balance is 1.238%. Hereinafter, using this database, an approach for analyzing "what is the characteristic of the customer whose periodicity balance changes frequently" will be described as an example.

[0027]

[Table 1]

FIG. 3 is an operation flowchart of the data analysis system shown in FIG. First, the database is read from the external storage device 24 into the intermediate processing device 22 (step 101), the compression parameter for determining the degree of feature preservation compression is input by the user (step 102), and the database is preserved and compressed according to the compression parameter. 1 to create a partial database (step 103). Step 1 when changing the degree of database compression
If no change is made, the process proceeds to step 105 (step 104).

Further, the user inputs the coordinate axes to be displayed three-dimensionally (step 105), and the records in the compressed partial database are three-dimensionally plotted and displayed as points on the three-dimensional space of the input coordinate axes (step 10).
6). If the display coordinate axis is changed, the process returns to step 105, and if not changed, the process proceeds to step 108 (step 107).

Further, the user designates two or more points three-dimensionally displayed on the screen with an input device such as a mouse, and displays a predetermined three-dimensional shape specified by these points (step 108). A record included in the three-dimensional shape is detected from the records in the partial database (step 109). The points representing the detected record are displayed separately from the points representing the other records (step 11).
0) If the three-dimensional shape is changed, the process returns to step 108, and if not changed, the process ends (step 11).
1).

However, in the operation of the data analysis system shown in FIG. 1, in addition to the series of operations from step 101 to step 111, (1) the scale of the database is small,
When there is no need for compression (proceeding from step 101 to step 105), (2) when the user sees the three-dimensional plot display, decides to change the number of display records and compresses again (step 107 to step 102). (3), (3) When the user decides to change the number of displayed records and compresses again at the stage when the user sees the record groups displayed in distinction in the three-dimensional space (step 111 to step 1).
(Return to 02), (4) When the user decides to change the display coordinate axis and inputs the display coordinate axis again at the stage when the user sees the record groups displayed in distinction in the three-dimensional space (step 111).
To step 105)) and so on.

FIG. 4 is a flowchart showing the feature preservation compression processing 1. FIG. 5 is a conceptual diagram of the feature storage compression processing 1.
Will be explained. For the attributes 1 and 2, 12 records are assumed to be distributed as shown in FIG.

In the characteristic preservation compression processing 1, first, all records are initialized to the non-selected state (step 201). One (x) is arbitrarily selected from the records in the non-selected state (step 202), and the distance D (x, y) between x and the record y in the non-selected state other than x is calculated (step 2).
03). That is, for example, when the record 301 is first selected, the distance between the record 301 and the other records is calculated.

Furthermore, the number N of records y whose distance is smaller than the threshold value is calculated (step 204), and the number of y and the number of records y are added to the partial database and output (step 205). And the record y whose distance is smaller than the threshold is set to the selected state (step 20).
6). That is, a record whose distance from the record 301 is smaller than a predetermined threshold value (30 in FIG. 5).
6, 307, 309, 311, 312) is the record 3
It is represented by 01 and is excluded from the subsequent record selection candidates. In the record 301, the number and the number of representative records (in FIG. 5, the number is 30
6, 307, 309, 311, 312, the number is 6
Items) and then add to the partial database. However, instead of the representative (compressed) record number added to the partial database, a customer code or other record identifying record may be used.

Furthermore, if all the records are not in the selected state, the process returns to step 202, and if all the records are in the selected state, the process ends (step 207).
That is, the same process is repeated thereafter, the record 308 is selected as the record number 2, the record 304 is selected as the record number 3 and the record 305 is selected as the record number 1, and the selection of the partial database is completed.

The distance is defined by, for example, D (x, y) = ΣW _i (x _i −y _i ) ² (Equation 1). i is a subscript indicating an attribute, x _i
Is the value of the attribute i of the record x, and W _i is the weight for the attribute i.

FIG. 6 is a flow chart showing the process of determining the threshold value used in the feature preservation compression process of the present invention. The factor that determines the compression ratio of the original database and the partial database after compression (the number of records of the partial database after compression / the number of records of the original database) is the threshold value, and in order to achieve the desired compression ratio. Needs to determine the threshold. The processing method for determining this threshold will be described below.

First, a small number (N) of all records are randomly sampled (step 401). An arbitrary record x is selected from N sampled records (step 402), and the distances to all unsampled records y are calculated (step 40).
3). Note that the above equation (Equation 1) is used for the distance.

Further, the records are arranged in order from the one with the smallest distance, and the number M ((the number of records of the original database / the number of records of the partial database after compression) is determined according to the degree of compression.
Is converted to an integer) and the distance MD to the (third) record is calculated (step 404). If the MD has not been calculated for all the N records sampled, the process returns to step 402, and if the MD has been calculated for all records, the process proceeds to step 406 (step 405). The average value of MDs of N sampled records is output as a threshold value, and the process ends (step 406).

Table 2 is an example of the partial database after the feature storage compression. The number of representatives and the representative customer code are added as attributes. For example, the record of the customer code “o31” is “ta29”, “ko55”, ...
On behalf of.

[0041]

[Table 2]

FIG. 7 shows an example of three-dimensional plot display.
The points representing each record are displayed in different patterns according to the number of representative records. In addition to the pattern, the number of representative records may be color, dot size, or shape.

8 to 10 show display examples when the user extracts a characteristic record group by using the three-dimensional shape display processing 3, the record detection processing 4, and the distinctive display processing 5. The user designated record A while viewing the three-dimensional plot display of the record (FIG. 8), and further designated record B (FIG. 9). At this time, as a three-dimensional shape,
Two points A and B are points on both ends of the diagonal line, and each side is X, Y,
A rectangular parallelepiped that is parallel to any of the Z axes is displayed.
Further, FIG. 10 shows a display example at the stage where the record C is designated by additionally designating some points. At this time, the record detection process 4 detects a record included in the inside of the three-dimensional shape (rectangular solid), and the distinction display process 5 distinguishes it from the other records and displays it by adding a cross.

FIG. 11 is a display example when the three-dimensional shape is a sphere, and a sphere having the record A as the center and the record B as a point on the sphere is displayed, and the records contained inside are displayed separately. Has been done.

According to the first embodiment described above, the user can determine the size (number) that can be discriminated by the three-dimensional plot display from the large-scale database with many attributes in the business field without impairing the characteristics of the entire database. Since a partial database can be created and displayed in a three-dimensional plot, the features of the entire database can be easily captured by visual judgment. In addition, when used in combination with a method such as statistical analysis, the user can actually visually check and evaluate the numerical result such as the tendency of the attribute of the database obtained by those methods.

A second embodiment, which is a modification of the first embodiment, will be described below. The record detected by the data detection process is saved as a new partial database. FIG. 12 is an operation flowchart of the data analysis system showing the second embodiment.

First, the database is read from the external storage device 24 into the intermediate processing device 22 (step 501), the compression parameter for determining the degree of feature preservation compression is input by the user (step 502), and the database is loaded according to the compression parameter. Feature storage compression 1 is performed to create a partial database (step 503). If the degree of compression of the database is changed, the process returns to step 502, and if not, the process proceeds to step 505 (step 5).
04).

A coordinate axis for three-dimensional display is input by the user (step 505), and the record in the compressed partial database is three-dimensionally plotted and displayed as a point on the three-dimensional space of the input coordinate axis (step 506). When the display coordinate axis is changed, the process returns to step 505, and when it is not changed, the process proceeds to step 508 (step 5
07). The user designates two or more points three-dimensionally displayed on the screen with an input device such as a mouse, and displays a predetermined three-dimensional shape specified by these points (step 508).

Further, records included in the three-dimensional shape are detected from the records in the partial database (step 509). The points representing the detected record are displayed separately from the points representing other records (step 51).
0) If the three-dimensional shape is changed, the process returns to step 508, and if not changed, the process proceeds to step 512 (step 511). The detected record is saved as a new partial database and the process ends (step 512).

However, the operation of the data analysis system shown in FIG. 12 is, in addition to the series of operations from step 501 to step 511, (1) when the scale of the database is small and compression is not necessary (step 501 to step (Step 505), (2) When the user views the three-dimensional plot display and decides to change the number of display records and compresses again (returning from step 507 to step 502), (3) user is three-dimensional When it is decided to change the number of displayed records at the stage of seeing the record groups displayed separately in the space and compressing again (steps 511 to 5)
(Return to 02), (4) when the user decides to change the display coordinate axis and inputs the display coordinate axis again at the stage when the user sees the record groups displayed in distinction in the three-dimensional space (step 511).
To step 505)) and the like.

It is also conceivable to newly read a new partial database storing the detected record and perform a series of processes (step 501 to step 512).

According to the second embodiment described above, the user analyzes a non-linear characteristic such as the bias of the distribution of records from a large-scale database with many attributes in the business field and selects a characteristic record group. It can be extracted and saved.

Since the representative record number is added to the compressed partial database, the representative record number added to the saved record group is used to extract the characteristic record group from the entire original database. You can search all. Furthermore, in the analysis using the three-dimensional shape display processing, the partial database detected and stored as a characteristic record group is used as a new database.
Local enlarged display can be realized by displaying a three-dimensional plot. Further, in the field of physical simulation, the same processing is performed on the three-dimensionally displayed data representing the fluid, so that a locally enlarged display is performed and a partial detailed analysis can be performed.

Further, in the three-dimensional display, only the analysis on three attributes can be performed at the same time, but the partial database detected and saved as a characteristic record group by the analysis using the three-dimensional shape display processing is used as a new database. By displaying a three-dimensional plot, analyzing on various axes, and further detecting and storing characteristic record groups, it is possible to realize multidimensional analysis of attributes. Furthermore, by applying feature saving and compression processing and using a reduced partial database, statistical methods, neuromodeling, rule induction methods, etc. were applied to reduce the processing time while preserving the characteristics of the entire database. Can analyze.

A third embodiment, which is a modification of the first embodiment, will be described below. The record detected by the data detection process is saved as a new partial database. FIG. 13 is an operation flowchart of the data analysis system showing the third embodiment.

First, the database is read from the external storage device 24 into the intermediate processing device 22 (step 601), the compression parameter for determining the degree of feature preservation compression is input by the user (step 602), and the database is stored according to the compression parameter. Feature storage compression 1 is performed to create a partial database (step 603). If the degree of compression of the database is changed, the process returns to step 602, and if not, the process proceeds to step 605 (step 6).
04).

Further, the coordinate axis for three-dimensional display is input by the user (step 605), and the record in the compressed partial database is three-dimensionally plotted and displayed as a point on the three-dimensional space of the input coordinate axis (step 60).
6). When the display coordinate axis is changed, the process returns to step 605, and when it is not changed, the process proceeds to step 608 (step 607). The user specifies two or more points that are three-dimensionally displayed on the screen with an input device such as a mouse.
A predetermined three-dimensional shape specified by these points is displayed (step 608).

Further, records included in the three-dimensional shape are detected from the records in the partial database (step 609). The points representing the detected record are displayed separately from the points representing other records (step 61).
0) If the three-dimensional shape is changed, the process returns to step 608, and if not changed, the process proceeds to step 612 (step 611). The detected record removed from the database is saved as a new partial database and the process ends (step 612).

However, the operation of the data analysis system shown in FIG. 13 is, in addition to the series of operations from step 601 to step 611, (1) when the database is small and compression is not necessary (step 601 to step 601). (Step 605), (2) When the user decides to change the number of displayed records at the stage of viewing the three-dimensional plot display and compresses again (returning from step 607 to step 602), (3) user three-dimensional When it is decided to change the number of displayed records at the stage of seeing the record groups displayed separately in the space and compressing again (steps 611 to 6)
(Return to 02), (4) when the user decides to change the display coordinate axis and inputs the display coordinate axis again at the stage when the user sees the record groups displayed separately in the three-dimensional space (step 611).
From step S605 to step 605).

It is also conceivable to newly read a new partial database in which the detected record is removed from the database and perform a series of processes (steps 601 to 612).

According to the third embodiment described above, the user analyzes a non-linear characteristic such as a bias in the distribution of records from a large-scale database with many attributes in the business field and records exceptional records from the database. The group can be removed and saved.

Since the representative record number is added to the compressed partial database, the representative record number added to the saved record group is used to make an exceptional record from the entire original database. You can search all record groups except. In the analysis using the three-dimensional shape display processing, the partial database saved by removing the exceptional record group from the database is displayed as a new database in three-dimensional plot display, analyzed in various axes, and further analyzed. By repeatedly detecting and removing a record group, exceptional record removal related to multidimensional attributes can be realized. Feature preservation compression processing 1
With regard to (2), the same effect can be expected not only for three-dimensional display but also for two-dimensional display.

[0063]

According to the data analysis system of the present invention, (1) feature storing / compressing means, three-dimensional plot displaying means, and 3
Since it has a three-dimensional shape display means, a record detection means, and a distinction display means, a partial database of a discriminable scale is generated from a database consisting of a large number of records while the characteristics such as the distribution state of the records are stored, and three-dimensional plotting is performed. It is possible to efficiently extract characteristic record groups while grasping the overall tendency by the display.

(2) The feature storing / compressing means randomly extracts records, calculates the distance between the extracted record and the other records, and records the distance of which is smaller than a predetermined threshold value. Since it is represented by the extracted records, it is excluded from the subsequent record extraction candidates and the extracted records are added to the partial database, so the characteristics such as the distribution status of the records are faithfully reproduced from the database consisting of a large number of records. You can create a partial database that is reflected in.

(3) Since the three-dimensional shape display means can specify a rectangular parallelepiped or a sphere determined by two specific two-dimensional points as the three-dimensional range, the desired record is not missed, and the characteristic is selected from the record group displayed in the three-dimensional plot. You can extract a record.

(4) Since the record detection means can detect and store the records within the range indicated by the specified three-dimensional shape, it is necessary to extract and store the characteristic record group from the database of a large number of records. You can

(5) Since the record detecting means can detect and delete the records within the range indicated by the specified three-dimensional shape, it is possible to remove the exceptional record group from the database consisting of a large number of records.

[Brief description of drawings]

FIG. 1 is a conceptual diagram of an information processing system showing a first embodiment of the present invention.

FIG. 2 is a hardware configuration diagram of the data analysis system of FIG.

FIG. 3 is an operation flowchart of the data analysis system shown in FIG.

FIG. 4 is a flowchart showing a feature preservation compression process 1.

FIG. 5 is a conceptual diagram of a feature preservation compression process 1.

FIG. 6 is a flowchart showing a process of determining a threshold value of the feature preservation compression process 1.

FIG. 7 is an example of a three-dimensional plot display.

FIG. 8 is a display example when a record A is designated in the three-dimensional shape display processing.

FIG. 9 is a display example when a record B is additionally designated in the three-dimensional shape display processing.

FIG. 10 shows a display example when a record C is additionally specified in the three-dimensional shape display process.

FIG. 11 is a display example when the three-dimensional shape is a sphere in the three-dimensional shape display processing.

FIG. 12 is an operation flowchart of the data analysis system showing the second embodiment.

FIG. 13 is an operation flowchart of the data analysis system showing the third embodiment.

[Explanation of symbols]

1: Feature preservation compression, 2: Three-dimensional plot display, 3: Three-dimensional shape display, 4: Record detection, 5: Discrimination display, 21:
Input device, 22: central processing unit, 23: display device, 2
4: External storage device.

 ─────────────────────────────────────────────────── ─── Continuation of the front page (72) Inventor Yukiyasu Ito 5030 Totsuka-cho, Totsuka-ku, Yokohama-shi, Kanagawa, Ltd., Ltd., Software Development Division, Hitachi, Ltd. Bachi Co., Ltd. Hitachi, Ltd. System Development Laboratory

Claims (6)

[Claims] 1. A data analysis system for deriving a relationship between attributes of records contained in a database consisting of a set of records, or predicting an output for an arbitrary input using the relationship, while preserving features. Feature storing / compressing means for compressing the database, three-dimensional plot displaying means for displaying the records as points on an arbitrary three-dimensional space, and a specific 2 on the space.
Three-dimensional shape display means for displaying a predetermined three-dimensional shape determined by one or more points in the space, record detection means for detecting records included in the three-dimensional shape, and the detected records are displayed separately. A data analysis system, comprising: a distinctive display means. 2. The feature storing / compressing means randomly extracts a record from a database, calculates a distance between the record and another record, and records a record whose distance is smaller than a predetermined threshold value. The data analysis system according to claim 1, wherein the extracted record is represented as a representative, and the extracted record is removed from subsequent record extraction candidates, and the extracted record is added to the partial database. 3. The predetermined three-dimensional shape is such that two specific points are points at both ends of a diagonal line, and each side is X, Y, Z.
The data analysis system according to claim 1, wherein the data analysis system is a rectangular parallelepiped parallel to any of the axes. 4. The data analysis system according to claim 1, wherein the predetermined three-dimensional shape is a sphere having a specific one point as a center and another point as a point on a spherical surface. 5. The data analysis system according to claim 1, wherein the record detection means stores the detected record as a new partial database. 6. The data analysis system according to claim 1, wherein the record detection means stores the detected record deleted from the database as a new partial database.