CN109460791A - A kind of arest neighbors method for detecting abnormality based on edge samples Density Metric - Google Patents
A kind of arest neighbors method for detecting abnormality based on edge samples Density Metric Download PDFInfo
- Publication number
- CN109460791A CN109460791A CN201811351192.5A CN201811351192A CN109460791A CN 109460791 A CN109460791 A CN 109460791A CN 201811351192 A CN201811351192 A CN 201811351192A CN 109460791 A CN109460791 A CN 109460791A
- Authority
- CN
- China
- Prior art keywords
- test point
- point
- training
- radius
- distance
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/22—Matching criteria, e.g. proximity measures
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
Landscapes
- Engineering & Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Theoretical Computer Science (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Artificial Intelligence (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Data Exchanges In Wide-Area Networks (AREA)
Abstract
The embodiment of the present invention proposes a kind of arest neighbors method for detecting abnormality based on edge samples Density Metric, it include: multiple sub- training sets that multiple stochastical sampling obtains normal sample, distance of each point away from its closest approach in subset is calculated in conjunction with Euclidean distance, region is constructed by radius of the distance, will not belong to the test point in any region as global abnormal；To the test point of non-global abnormal, its nearest training points and the nearest training points of the training points are found, using the ratio of two o'clock region radius as the global metric of test point exception；Using the ratio of the minimum distance of test point to its nearest training points edges of regions and the zone radius as the Local Metric value of the point exception, the isolation score of test point is obtained in conjunction with metric twice, using the average value of multiple sub- Quarantine at assembly sites scores as abnormality score.Technical solution provided in an embodiment of the present invention has fully considered the distribution characteristics of edge samples, can effectively solve local anomaly test problems in edge samples adjacent domain.
Description
[technical field]
The present invention relates to machine learning field method for detecting abnormality more particularly to a kind of based on edge samples Density Metric
Arest neighbors method for detecting abnormality.
[background technique]
When solving two classification problems using machine learning method, it is extremely unbalanced existing that there are data set distribution between class
As that is, the quantity of exception class sample far less than normal class sample size or does not have the phenomenon that exception class sample.For the former
The case where there are exception class samples, common technology is broadly divided into the method based on data and the method based on algorithm at present.It is based on
The method of data, which refers to, is reconstructed data set itself by over-sampling or lack sampling, so that unbalanced sample distribution becomes
Compare balance.Oversampler method is by sampling with replacement, generates the methods of the similar sample of exception class and increase exception class sample
This number, to achieve the purpose that equilibrium data collection, which will increase the training time, improves the complexity of calculating, easily leads
Cause over-fitting.Lack sampling method is the number of normal class sample to be reduced, to reduce by the normal class sample of discard portion
The unbalanced degree of data set, which can lose the useful information of normal class sample, and have certain blindness.It is based on
The method of algorithm refers to the extreme imbalance problem of distribution between class for data sets, to have supervision algorithm do it is suitably modified or use
Unsupervised algorithm, to have supervision algorithm modify common method include introduce the cost-sensitive factor, to the minority of classification error
Class sample is punished, such as cost-sensitive decision tree, Cost Sensitive Support Vector Machines.Or it is carried out using a series of classifiers
Study, and each learning outcome integrate to obtaining learning effect more better than single classifier, as AdaBoost,
Bagging etc..But these algorithms can not really solve sample number without fundamentally changing the disequilibrium between data set class
According to the classification problem under extreme distribution occasion.Especially in the case where no exception class sample, the above method based on data and
Based on have supervision algorithm method not can effectively solve the problem.Therefore, for such situation, can consider using unsupervised algorithm
Such issues that solve, to be showed in such algorithm preferably wherein isolating forest algorithm.Isolated forest algorithm use every time one with
Machine hyperplane carrys out the every sub-spaces generated after cutting data space and its cutting, only one number inside every subspace
Strong point or until reaching preset termination condition.The algorithm can only utilize normal sample, more effectively handle sample number
According to the classification problem under extreme distribution occasion, still, isolated forest algorithm can not detect local anomaly well, including positive and negative
The exception of sample cross and the exception circular by normal class sample.In response to this, scholar proposes based on isolation
Nearest neighbor algorithm.The algorithm only carries out stochastical sampling in normal sample data set, by establishing area of isolation, calculates test point
Relative position between nearest training points judges the intensity of anomaly of the test point.But this algorithm is for surveying
The judgement of pilot exception depends critically upon the size of area of isolation constructed by its arest neighbors training points.Training sample edge compared with
Under conditions of sparse, the area of isolation of training sample is larger, may will affect the judgement to test point intensity of anomaly, therefore need
Further measurement is done to test point intensity of anomaly inside edge samples area of isolation.
[summary of the invention]
In view of this, the embodiment of the present invention proposes a kind of arest neighbors abnormality detection side based on edge samples Density Metric
Method, to solve local anomaly test problems in edge samples adjacent domain.
The embodiment of the present invention proposes a kind of arest neighbors method for detecting abnormality based on edge samples Density Metric, comprising:
Multiple stochastical sampling obtains multiple sub- training sets of normal sample, in conjunction with each point in Euclidean distance calculating subset away from it
The distance of closest approach constructs region by radius of the distance, will not belong to the test point in any region as global abnormal；
To the test point of non-global abnormal, its nearest training points and the nearest training points of the training points are found, by two o'clock institute
In global metric of the ratio as test point exception of zone radius；
The minimum distance of test point to its nearest training points edges of regions and the ratio of the zone radius is different as the point
Normal Local Metric value obtains the isolation score of test point in conjunction with metric twice, by being averaged for multiple sub- Quarantine at assembly sites scores
Value is used as abnormality score.
In the above method, multiple stochastical sampling obtains multiple sub- training sets of normal sample, calculates son in conjunction with Euclidean distance
Distance of each point away from its closest approach is concentrated, region is constructed by radius of the distance, will not belong to the test point conduct in any region
The method of global abnormal are as follows: multiple simple random sampling is carried out to normal sample data set D, obtaining multiple sample sizes is ψ's
Sub- training set Si, i is integer and 1≤i≤t, t are the number of subset, appropriate value can be selected according to the actual situation, in every height
Training set SiIn, the distance between each point is calculated based on Euclidean distance, using each training points a as regional center, with a to it most
Nearly training points ηaDistance τ (a) as zone radius construct a region, point a is isolated with training points other in subset, wherein
a,ηa∈Si, have for the radius distance τ (a) of point a defined below:
τ (a)=| | a- ηa| | formula (1)
For every sub- training set SiIf c is the training points nearest away from test point x, c ∈ Si, then whether test point x is complete
Office is abnormal to be defined by following rule:
When τ (c) < τ (x), x is global abnormal；
According to formula (1), τ (x) and τ (c) is the radius distance of point x He point c respectively,It is whether determining x is global
Abnormal line of demarcation.
In the above method, to the test point of non-global abnormal, its nearest training points and the nearest training of the training points are found
Point, using the ratio of two o'clock region radius as the method for the global metric of test point exception are as follows: instructed for every height
Practice collection SiIf b is SiIn any one training points, will using b as the centre of sphere, τ (b) be radius building suprasphere be denoted as B (b), then
To any one training points y in B (b), just like giving a definition:
Y:| | y-b | | < τ (b) formula (2)
For not being the test point of global abnormal, if c is the training points nearest away from test point x, ηcIt is the training nearest away from c
Point, c, ηc∈Si, according to formula (2), B (ηc) and B (c) be with η respectivelycIt is the centre of sphere with c, with τ (ηc) and τ (c) be radius hypersphere
Body, B (ηc) and B (c) radius ratioIt is that training points c is measured relative to the isolation of its neighborhood, it willAs test
The global metric of point x intensity of anomaly,It is bigger, show B (ηc) and B (c) relative radius gap
Smaller, test point intensity of anomaly is lower；Conversely,It is smaller, show B (ηc) and the relative radius gap of B (c) it is bigger, test
Point intensity of anomaly is higher.
In the above method, by test point to the minimum distance of its nearest training points edges of regions and the ratio of the zone radius
As the Local Metric value of the point exception, the isolation score of test point is obtained in conjunction with metric twice, by multiple sub- Quarantine at assembly sites
Method of the average value of score as abnormality score are as follows: for each test point x, set up an office c ∈ SiIt is the training nearest away from x
Point, then it is internal in B (c), to the minimum distance d of test point x to B (c) edges of regions, just like giving a definition:
D=τ (c)-| | c-x | | formula (3)
Wherein, τ (c) is the radius distance of training points c, willAs the Local Metric value of test point x intensity of anomaly,AndIt is bigger, show that test point x is closer away from its nearest training points c, intensity of anomaly is lower；Conversely,More
It is small, show that test point x is remoter away from its nearest training points c, intensity of anomaly is higher；For every sub- training set Si, test point x every
It is defined as follows from score I (x):
Multiple repairing weld is carried out to normal sample data set and obtains multiple sub- training set { S1,S2,...,St, t is of subset
Number, can select appropriate value, respectively in each subset S according to the actual situationiThe isolation point of test point x is calculated in (1≤i≤t)
Number, then for the abnormality score of test point xJust like giving a definition:
Wherein IiIt (x) is isolation score of the test point x in i-th of subset；Abnormality scoreThe exception of x can be measured
Degree,It is more big, show that test point x intensity of anomaly is higher；Conversely,It is smaller, show that test point x intensity of anomaly is got over
It is low.
[Detailed description of the invention]
In order to illustrate the technical solution of the embodiments of the present invention more clearly, below will be to needed in the embodiment attached
Figure is briefly described, it should be apparent that, drawings in the following description are only some embodiments of the invention, for this field
For those of ordinary skill, without any creative labor, it can also be obtained according to these attached drawings other attached
Figure.
Fig. 1 is the stream for the arest neighbors method for detecting abnormality based on edge samples Density Metric that the embodiment of the present invention is proposed
Journey schematic diagram；
Fig. 2 is that the arest neighbors method for detecting abnormality based on edge samples Density Metric that the embodiment of the present invention is proposed calculates
The schematic diagram of test point isolation score.
[specific embodiment]
For a better understanding of the technical solution of the present invention, being retouched in detail to the embodiment of the present invention with reference to the accompanying drawing
It states.
It will be appreciated that described embodiments are only a part of the embodiments of the present invention, instead of all the embodiments.Base
Embodiment in the present invention, it is obtained by those of ordinary skill in the art without making creative efforts it is all its
Its embodiment, shall fall within the protection scope of the present invention.
The embodiment of the present invention provides the arest neighbors method for detecting abnormality based on edge samples Density Metric, as shown in Figure 1, its
It, should by the flow diagram for the arest neighbors method for detecting abnormality based on edge samples Density Metric that the embodiment of the present invention proposes
Method the following steps are included:
Step 101, multiple stochastical sampling obtains multiple sub- training sets of normal sample, calculates in subset in conjunction with Euclidean distance
Distance of each point away from its closest approach constructs region by radius of the distance, will not belong to the test point in any region as global
It is abnormal.
Specifically, carrying out multiple simple random sampling to normal sample data set D, the son that multiple sample sizes are ψ is obtained
Training set Si, i is integer and 1≤i≤t, t are the number of subset, can select appropriate value according to the actual situation, is instructed in every height
Practice collection SiIn, the distance between each point is calculated based on Euclidean distance, it is nearest to it with a using each training points a as regional center
Training points ηaDistance τ (a) as zone radius construct a region, point a is isolated with training points other in subset, wherein a,
ηa∈Si, have for the radius distance τ (a) of point a defined below:
τ (a)=| | a- ηa| | formula (1)
For every sub- training set SiIf c is the training points nearest away from test point x, c ∈ Si, then whether test point x is complete
Office is abnormal to be defined by following rule:
When τ (c) < τ (x), x is global abnormal；
According to formula (1), τ (x) and τ (c) is the radius distance of point x He point c respectively,It is whether determining x is global
Abnormal line of demarcation.
Step 102, to the test point of non-global abnormal, its nearest training points and the nearest training points of the training points are found,
Using the ratio of two o'clock region radius as the global metric of test point exception.
Specifically, for every sub- training set SiIf b is SiIn any one training points, will be using b as the centre of sphere, τ (b) is
The suprasphere of radius building is denoted as B (b), then to any one training points y in B (b), just like giving a definition:
Y:| | y-b | | < τ (b) formula (2)
For not being the test point of global abnormal, if c is the training points nearest away from test point x, ηcIt is the training nearest away from c
Point, c, ηc∈Si, according to formula (2), B (ηc) and B (c) be with η respectivelycIt is the centre of sphere with c, with τ (ηc) and τ (c) be radius hypersphere
Body, B (ηc) and B (c) radius ratioIt is that training points c is measured relative to the isolation of its neighborhood, it willAs test
The global metric of point x intensity of anomaly,It is bigger, show B (ηc) and B (c) relative radius gap
Smaller, test point intensity of anomaly is lower；Conversely,It is smaller, show B (ηc) and the relative radius gap of B (c) it is bigger, test
Point intensity of anomaly is higher.
Step 103, the ratio of the minimum distance of test point to its nearest training points edges of regions and the zone radius is made
For the Local Metric value of the point exception, the isolation score of test point is obtained in conjunction with metric twice, by multiple sub- Quarantine at assembly sites point
Several average values are as abnormality score.
Specifically, set up an office c ∈ S for each test point xiIt is the training points nearest away from x, then it is internal in B (c), to survey
The minimum distance d of pilot x to B (c) edges of regions, just like giving a definition:
D=τ (c)-| | c-x | | formula (3)
Wherein, τ (c) is the radius distance of training points c, willAs the Local Metric value of test point x intensity of anomaly,AndIt is bigger, show that test point x is closer away from its nearest training points c, intensity of anomaly is lower；Conversely,More
It is small, show that test point x is remoter away from its nearest training points c, intensity of anomaly is higher；For every sub- training set Si, test point x every
It is defined as follows from score I (x):
Multiple repairing weld is carried out to normal sample data set and obtains multiple sub- training set { S1,S2,...,St, t is of subset
Number, can select appropriate value, respectively in each subset S according to the actual situationiThe isolation point of test point x is calculated in (1≤i≤t)
Number, then for the abnormality score of test point xJust like giving a definition:
Wherein IiIt (x) is isolation score of the test point x in i-th of subset；Abnormality scoreThe exception of x can be measured
Degree,It is more big, show that test point x intensity of anomaly is higher；Conversely,It is smaller, show that test point x intensity of anomaly is got over
It is low.
The arest neighbors method for detecting abnormality based on edge samples Density Metric that algorithm 1 is proposed by the embodiment of the present invention
Pseudocode:
Fig. 2 is the schematic diagram that this method calculates test point isolation score, and wherein x is test point, and c is nearest away from test point x
Training points, ηcFor the training points nearest away from training points c, great circle region and roundlet region respectively indicate c and ηcArea of isolation B
(c) and B (ηc), d is the minimum distance of the edges of regions of test point x to its nearest training points c.
Table solves 10 first is that the embodiment of the present invention provides the arest neighbors method for detecting abnormality based on edge samples Density Metric
When group public data collection classification task, title, sample size, dimension, exceptional sample ratio and the two methods of each data set
The contrast and experiment of recall rate, wherein control methods is the nearest neighbor algorithm based on isolation in the embodiment of the present invention
(isolation using Nearest Neighbour Ensembles, iNNE), based on the nearest of edge samples Density Metric
Adjacent method for detecting abnormality is denoted as iNNE_add.As shown in Table 1, method proposed by the invention public data concentrate compared to
Control methods averagely improves 1.42% in recall rate.Particularly, lifting values of the method on Waveform data set are proposed
Highest reaches 3.0%.The method that the embodiment of the present invention is proposed can fully consider the distribution characteristics of edge samples, can effectively solve
Certainly local anomaly test problems in edge samples adjacent domain.
Table one
| Dataset name | Sample size | Dimension | Exceptional sample ratio (%) | INNE recall rate (%) | INNE_add recall rate (%) |
| Skin (Skin) | 9521 | 3 | 3.7 | 78.2 | 80.3 |
| Pima (Pima) | 768 | 8 | 34.9 | 51.3 | 51.8 |
| Airliner (Shuttle) | 49097 | 9 | 7.2 | 99.4 | 99.5 |
| Glass (Glass) | 214 | 9 | 4.2 | 66.6 | 69.3 |
| Grape wine (Wine) | 5318 | 11 | 4.5 | 90.9 | 92.1 |
| Disk (Disk) | 35578 | 14 | 1.1 | 63.4 | 65.5 |
| Waveform (Waveform) | 3505 | 21 | 4.6 | 61.1 | 64.1 |
| Satellite (Satellite) | 6435 | 36 | 31.6 | 53.1 | 54.7 |
| Arrhythmia cordis (Arrhythmia) | 452 | 274 | 14.6 | 70.3 | 70.6 |
| Harbour (Har) | 5195 | 561 | 10.1 | 53.2 | 53.8 |
In conclusion the embodiment of the present invention has the advantages that
In the technical solution that the present invention is implemented, multiple stochastical sampling obtains multiple sub- training sets of normal sample, in conjunction with Europe
Family name's distance calculates distance of each point away from its closest approach in subset, and region is constructed by radius of the distance, will not belong to any region
Test point as global abnormal；To the test point of non-global abnormal, its nearest training points and the nearest instruction of the training points are found
Practice point, using the ratio of two o'clock region radius as the global metric of test point exception；Test point is instructed recently to it
Practice the minimum distance of point edges of regions and Local Metric value of the ratio as the point exception of the zone radius, in conjunction with measuring twice
Value obtains the isolation score of test point, using the average value of multiple sub- Quarantine at assembly sites scores as abnormality score.The embodiment of the present invention
The technical solution of offer has fully considered the distribution characteristics of edge samples, can effectively solve part in edge samples adjacent domain
Abnormality detection problem.
The foregoing is merely illustrative of the preferred embodiments of the present invention, is not intended to limit the invention, all in essence of the invention
Within mind and principle, any modification, equivalent substitution, improvement and etc. done be should be included within the scope of the present invention.
Claims (4)
1. a kind of arest neighbors method for detecting abnormality based on edge samples Density Metric, the method step include:
(1) multiple stochastical sampling obtains multiple sub- training sets of normal sample, in conjunction with each point in Euclidean distance calculating subset away from it
The distance of closest approach constructs region by radius of the distance, will not belong to the test point in any region as global abnormal；
(2) to the test point of non-global abnormal, its nearest training points and the nearest training points of the training points are found, it will be where two o'clock
Global metric of the ratio of zone radius as test point exception；
(3) using the ratio of the minimum distance of test point to its nearest training points edges of regions and the zone radius as point exception
Local Metric value, the isolation score of test point is obtained in conjunction with metric twice, by the average value of multiple sub- Quarantine at assembly sites scores
As abnormality score.
2. the method according to claim 1, wherein repeatedly stochastical sampling obtains multiple son training of normal sample
Collection calculates distance of each point away from its closest approach in subset in conjunction with Euclidean distance, region is constructed by radius of the distance, will not belong to
The test point in any region is described as follows as global abnormal: being carried out multiple simple randomization to normal sample data set D and is adopted
Sample obtains the sub- training set S that multiple sample sizes are ψi, i is integer and 1≤i≤t, t are the number of subset, can be according to reality
Situation selects appropriate value, in every sub- training set SiIn, the distance between each point is calculated based on Euclidean distance, by each training
Point a is as regional center, with a to its nearest training points ηaDistance τ (a) as zone radius construct a region, make point a with
Other training points are isolated in subset, wherein a, ηa∈Si, have for the radius distance τ (a) of point a defined below:
τ (a)=| | a- ηa| | formula (1)
For every sub- training set SiIf c is the training points nearest away from test point x, c ∈ Si, then whether test point x is global different
Often defined by following rule:
When τ (c) < τ (x), x is global abnormal；
According to formula (1), τ (x) and τ (c) is the radius distance of point x He point c respectively,It is whether determining x is global abnormal
Line of demarcation.
3. the method according to claim 1, wherein finding its training recently to the test point of non-global abnormal
The nearest training points of point and the training points, using the global measurement that the ratio of two o'clock region radius is abnormal as the test point
Value, illustrates are as follows: for every sub- training set SiIf b is SiIn any one training points, will be using b as the centre of sphere, τ (b) is
The suprasphere of radius building is denoted as B (b), then to any one training points y in B (b), just like giving a definition:
Y:| | y-b | | < τ (b) formula (2)
For not being the test point of global abnormal, if c is the training points nearest away from test point x, ηcIt is the training points nearest away from c, c,
ηc∈Si, according to formula (2), B (ηc) and B (c) be with η respectivelycIt is the centre of sphere with c, with τ (ηc) and τ (c) be radius suprasphere, B
(ηc) and B (c) radius ratioIt is that training points c is measured relative to the isolation of its neighborhood, it willIt is different as test point x
The global metric of Chang Chengdu,It is bigger, show B (ηc) and the relative radius gap of B (c) it is smaller,
Test point intensity of anomaly is lower；Conversely,It is smaller, show B (ηc) and the relative radius gap of B (c) it is bigger, test point is different
Chang Chengdu is higher.
4. the method according to claim 1, wherein by test point to the nearest of its nearest training points edges of regions
Local Metric value of the ratio of distance and the zone radius as the point exception, obtains the isolation of test point in conjunction with metric twice
Score is illustrated using the average value of multiple sub- Quarantine at assembly sites scores as abnormality score are as follows: for each test point x, if
Point c ∈ SiIt is the training points nearest away from x, then it is internal in B (c), to the minimum distance d of test point x to B (c) edges of regions, just like
Give a definition:
D=τ (c)-| | c-x | | formula (3)
Wherein, τ (c) is the radius distance of training points c, willAs the Local Metric value of test point x intensity of anomaly,AndIt is bigger, show that test point x is closer away from its nearest training points c, intensity of anomaly is lower；Conversely,More
It is small, show that test point x is remoter away from its nearest training points c, intensity of anomaly is higher；For every sub- training set Si, test point x every
It is defined as follows from score I (x):
Multiple repairing weld is carried out to normal sample data set and obtains multiple sub- training set { S1,S2,...,St, t is the number of subset,
Appropriate value can be selected according to the actual situation, respectively in each subset SiThe isolation score of test point x is calculated in (1≤i≤t),
Then for the abnormality score of test point xJust like giving a definition:
Wherein IiIt (x) is isolation score of the test point x in i-th of subset；Abnormality scoreThe intensity of anomaly of x can be measured,It is more big, show that test point x intensity of anomaly is higher；Conversely,It is smaller, show that test point x intensity of anomaly is lower.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811351192.5A CN109460791A (en) | 2018-11-14 | 2018-11-14 | A kind of arest neighbors method for detecting abnormality based on edge samples Density Metric |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811351192.5A CN109460791A (en) | 2018-11-14 | 2018-11-14 | A kind of arest neighbors method for detecting abnormality based on edge samples Density Metric |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN109460791A true CN109460791A (en) | 2019-03-12 |
Family
ID=65610362
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201811351192.5A Pending CN109460791A (en) | 2018-11-14 | 2018-11-14 | A kind of arest neighbors method for detecting abnormality based on edge samples Density Metric |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109460791A (en) |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112329350A (en) * | 2020-11-17 | 2021-02-05 | 南京航空航天大学 | Airplane lead-acid storage battery abnormity detection semi-supervision method based on isolation |
| CN112562771A (en) * | 2020-12-25 | 2021-03-26 | 北京邮电大学 | Disk anomaly detection method based on neighborhood partition and isolation reconstruction |
| CN117131454A (en) * | 2023-10-23 | 2023-11-28 | 北京华力兴科技发展有限责任公司 | X-ray thickness measurement abnormal data monitoring method |
-
2018
- 2018-11-14 CN CN201811351192.5A patent/CN109460791A/en active Pending
Cited By (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112329350A (en) * | 2020-11-17 | 2021-02-05 | 南京航空航天大学 | Airplane lead-acid storage battery abnormity detection semi-supervision method based on isolation |
| CN112562771A (en) * | 2020-12-25 | 2021-03-26 | 北京邮电大学 | Disk anomaly detection method based on neighborhood partition and isolation reconstruction |
| CN112562771B (en) * | 2020-12-25 | 2022-07-26 | 北京邮电大学 | Disk anomaly detection method based on neighborhood partition and isolation reconstruction |
| CN117131454A (en) * | 2023-10-23 | 2023-11-28 | 北京华力兴科技发展有限责任公司 | X-ray thickness measurement abnormal data monitoring method |
| CN117131454B (en) * | 2023-10-23 | 2024-01-12 | 北京华力兴科技发展有限责任公司 | X-ray thickness measurement abnormal data monitoring method |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Chung et al. | Slice finder: Automated data slicing for model validation | |
| CN109460791A (en) | A kind of arest neighbors method for detecting abnormality based on edge samples Density Metric | |
| CN104887224B (en) | Feature extraction and automatic identifying method towards epileptic EEG Signal | |
| CN104330721B (en) | IC Hardware Trojan detecting method and system | |
| CN103136504B (en) | Face identification method and device | |
| CN105574547B (en) | Adapt to integrated learning approach and device that dynamic adjusts base classifier weight | |
| CN105069470A (en) | Classification model training method and device | |
| CN103048273B (en) | Fruit near infrared spectrum sorting method based on fuzzy clustering | |
| CN106126918B (en) | A kind of geographical space abnormal aggregation domain scanning statistical method based on interaction force | |
| CN104731916A (en) | Optimizing initial center K-means clustering method based on density in data mining | |
| KR101693405B1 (en) | Apparatus and method for detecting anomaly intrusion using local deviation factor graph based algorithm | |
| CN106482967A (en) | A kind of Cost Sensitive Support Vector Machines locomotive wheel detecting system and method | |
| CN109960808A (en) | A kind of text recognition method, device, equipment and computer readable storage medium | |
| CN111723897A (en) | Multi-modal feature selection method based on particle swarm optimization | |
| CN109976308A (en) | A kind of extracting method of the fault signature based on Laplce's score value and AP cluster | |
| CN109886284A (en) | Fraud detection method and system based on hierarchical clustering | |
| CN104809393A (en) | Shilling attack detection algorithm based on popularity classification features | |
| CN104765879A (en) | Density-based partitioning and clustering method for K center points in data mining | |
| CN113781483B (en) | Industrial product appearance defect detection method and device | |
| CN113191359A (en) | Small sample target detection method and system based on support and query samples | |
| CN103810482A (en) | Multi-information fusion classification and identification method | |
| Kamaladevi et al. | Tomek link undersampling with stacked ensemble classifier for imbalanced data classification | |
| CN113360392A (en) | Cross-project software defect prediction method and device | |
| CN111681710A (en) | Cell classification method and device based on gene expression characteristics and electronic equipment | |
| CN115935160A (en) | Air quality data processing method based on neighborhood rough set attribute reduction |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| RJ01 | Rejection of invention patent application after publication | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20190312 |