Unsupervised Artificial Neural Networks for Outlier Detection in High-Dimensional Data

Supplementary Material

Welcome to the supplementary material for the paper:

Daniel Popovic, Edouard Fouché and Klemens Böhm. 2019. Unsupervised Artificial Neural Networks for Outlier Detection in High-Dimensional Data. In 23rd European Conference on Advances in Databases and Information Systems (ADBIS' 19), September 8-11, 2019, Bled, Slovenia. Springer International Publishing. https://doi.org/10.1007/978-3-030-28730-6_1

This repository contains the reference implementation of the three approaches Autoencoder, Restricted Boltzmann Machine, and Self-Organising Map for outlier detection with details to reproduce the experiments in the paper. In addition, it contains all our experimental results, which are only partially shown in the paper due to space restrictions.

This repository is released under the AGPLv3 license. Please see the LICENSE file for details.

If you use this repository, please cite our paper:

@InProceedings{10.1007/978-3-030-28730-6_1,
author="Popovic, Daniel and Fouch{\'e}, Edouard and B{\"o}hm, Klemens",
title="Unsupervised Artificial Neural Networks for Outlier Detection in High-Dimensional Data",
booktitle="Advances in Databases and Information Systems",
year="2019",
publisher="Springer International Publishing",
pages="3--19",
isbn="978-3-030-28730-6",
doi="10.1007/978-3-030-28730-6_1"
}

Quick Start

Prerequisites

Download this repository to your local machine. It is recommended to use virtual environment to manage Python dependencies.

$ pip3 install -r requirements.txt

To re-enact the experiments, download the data sets as ZIP files at Datasets Download and unpack them.

Run the Outlier Detector

The outlier detector runs in four different modes, using either Autoencoder, Restricted Boltzmann Machine, or Self-Organising Map alone, or run all three methods together. The detector supports two globally used optional parameters:

-w: Writes the outlier detection results into CSV files. For each outlier detection method, a separate file is saved. The files contain an incremented data object ID, the real outlier label provided in the data set, and the method's respective outlier score.
-d: The column delimiter for data sets in CSV format. The default delimiter is ;.

Autoencoder

To run the Autoencoder alone, use the command line positional parameter ae to specify the method, and as the second positional parameter, the path and file name of the data set.

$ python detector.py ae mypath/mydataset.arff

For the Autoencoder, the following optional parameters exist:

-a: The number of training epochs. Default is 20.
-e: The encoding factor. Default is 0.8.

Restricted Boltzmann Machine

To run the Restricted Boltzmann Machine alone, use the positional parameter rbm to specify the method, and the data set path and file name as a second positional parameter.

For the Restricted Boltzmann Machine, the following optional parameters exist:

-r: The number of training epochs. Default is 100.
-n: The proportion of hidden neurons. Default is 0.8.

Self-Organising Map

To run the Self-Organising Map alone, use the positional parameter som to specify the method, and the data set path and file name as a second positional parameter.

$ python detector.py som mypath/mydataset.arff

For the Self-Organising Map, the following optional parameters exist:

-s: The number of training epochs. Default is 100.
-g: The grid size or the number of neuron rows and columns in the quadratic SOM grid. Default is 2.

All Methods

To run all three methods at once, use the positional parameter all to specify all three methods, and the data set path and file name as the second positional parameter.

$ python detector.py all mypath/mydataset.arff

Each optional parameter that exists for one of the three methods (see above) is also valid to use with all three methods at once. If an optional parameter is not specified, the default value is used.

Data Set Structure

The outlier detector supports three file types for data sets: ARFF files (.arff), CSV files (.csv), and MATLAB files < version 7.3 (.mat).

ARFF Files

ARFF files are expected to have a number of feature attribute definitions and a label attribute definition at the end, which can either be the attribute 'class' with possible values {0,1} or the attribute 'outlier' with possible values {'no','yes'}

Example (with 'class' attribute):

@RELATION musk

@ATTRIBUTE 'att1' real
...
@ATTRIBUTE 'att166' real
@ATTRIBUTE 'class' {0,1} 

@DATA
46.,-108.,60.,...,-50.,-112.,96.,1
...
43.,64.,41.,...,-69.,-270.,-6.,0

Example (with 'outlier' attribute):

@RELATION musk

@ATTRIBUTE 'att1' real
...
@ATTRIBUTE 'att166' real
@ATTRIBUTE 'outlier' {'no','yes'} 

@DATA
46.,-108.,60.,...,-50.,-112.,96.,'yes'
...
43.,64.,41.,...,-69.,-270.,-6.,'no'

CSV Files

CSV files are expected to consist of of character-separated rows of features, with the label as the row's last value, and without a header row. The default column delimiter is ;, but any other reasonable delimiter can be used as passed as parameter delimiter to the detector. Example:

46.;-108.;60.;...;-50.;-112.;96.;1.
...
43.;64.;41.;...;-69.;-270.;-6.;0.

MATLAB Files

MATLAB .mat files (< version 7.3) are expected to have the following structure:

X: Feature matrix
y: Label vector

Example (.mat file loaded with scipy.io.loadmat):

{'__header__': b'MATLAB 5.0 MAT-file, written by Octave 3.8.0, 2015-07-05 13:16:13 UTC',
 '__version__': '1.0',
 '__globals__': [],
 'X': array([[  46., -108.,  -60., ...,  -50., -112.,   96.],
        ...,
        [  43.,   64.,   41., ...,  -69., -270.,   -6.]]),
 'y': array([[1.],
        ...,
        [0.]])}

Implementation Release

Project Structure

Root Folder

Folders:

nnohd: The implementation of the unsupervised neural networks for outlier detection.
images: Contains images for this README.
datasets: The data sets used for evaluation.

Files:

README.md
LICENSE: The license file. This repository is published under GNU AFFERO GENERAL PUBLIC LICENSE Version 3.
requirements.txt: Refers to app/requirements.txt.

Application Structure

nnohd/
├── requirements.txt
├── ae.py
├── rbm.py
├── som.py
├── utils
│   └── preprocessing.py

Dependencies

Python 3.6+
Keras 2.2.4
Matplotlib 2.2.2
NumPy 1.16.2
Pandas 0.24.2
Scikit-learn 0.20.3
SciPy 1.2.1
Somoclu 1.7.5
Tensorflow 1.13.1

Additional Material

This section contains the complete data for the sections of the paper that are only partially filled due to space limitations.

Evaluated Data Sets

In this section we list the complete set of evaluated data sets. The evaluated high-dimensional data sets can be found as ZIP files at Datasets Download.

Show Table

Data Set	Dimensions	Data Objects	Outliers	Outlier Ratio
Arrhythmia-2	259	248	4	1.61%
Arrhythmia-5	259	256	12	4.69%
Arrhythmia-10	259	271	27	9.96%
Arrhythmia-20	259	305	61	20.00%
Arrhythmia-46	259	450	206	45.78%
Cardio	21	1,831	176	9.61%
Ecoli	7	336	9	2.68%
InternetAds-2	1,555	1,630	32	1.96%
InternetAds-5	1,555	1,682	84	4.99%
InternetAds-10	1,555	1,775	177	9.97%
InternetAds-19	1,555	1,966	368	18.72%
ISOLET-2	617	2,449	50	2.00%
ISOLET-4	617	2,499	100	4.00%
ISOLET-8	617	2,599	200	7.70%
ISOLET-14	617	2,799	400	14.29%
ISOLET-20	617	2,999	600	20.00%
Lympho	18	148	6	4.05%
MNIST	100	7,603	700	9.21%
Musk	166	3,062	97	3.12%
Optdigits	64	5,216	150	2.88%
P53	5,408	16,592	143	0.86%
Pendigits	16	6,870	156	2.27%
Seismic	11	2,584	170	6.58%
Thyroid	6	3,772	93	2.4%
Waveform	21	3,509	166	4.73%
Yeast	8	1,364	65	4.77%

Dataset Evaluation

ROC AUC for High Dimensional Data Sets

Show Table

Data Set	AE	SOM	RBM	HiCS	LOF	FastABOD	LoOP	OC-SVM	KNN
Arrhythmia-2	80.22	76.33	49.04	50.56	76.74	76.84	75.00	77.66	71.88
Arrhythmia-5	78.16	81.49	50.36	58.81	80.46	49.51	79.99	81.47	68.72
Arrhythmia-10	83.84	83.74	48.23	65.12	83.45	83.35	84.24	83.26	76.20
Arrhythmia-20	71.01	71.33	49.93	50.71	70.47	71.14	71.70	72.38	67.52
Arrhythmia-46	74.93	74.79	46.99	58.96	74.41	74.17	74.10	74.13	70.72
InternetAds-2	43.36	66.12	46.93	99.84	71.64	76.49	77.14	64.18	81.23
InternetAds-5	40.14	73.36	50.27	99.87	78.63	78.75	82.56	74.76	69.35
InternetAds-10	39.38	69.42	49.75	79.25	74.71	74.41	77.73	71.25	65.21
InternetAds-19	36.32	69.69	50.14	47.33	74.01	72.40	69.78	62.43	59.54
ISOLET-2	96.85	99.28	92.28	79.71	99.58	93.09	98.23	92.05	94.66
ISOLET-4	94.72	98.49	87.24	79.79	99.34	86.44	95.08	89.75	83.89
ISOLET-8	95.16	98.29	87.19	78.14	83.81	75.37	68.71	89.19	82.90
ISOLET-14	92.67	94.96	85.50	78.16	65.61	75.60	62.16	81.61	80.00
ISOLET-20	90.16	87.13	87.19	77.82	60.04	73.28	55.00	77.06	77.40
MNIST	82.06	81.07	49.87	51.74	80.34	54.35	71.66	76.46	72.74
Musk	100.00	100.00	95.60	99.60	84.00	5.11	51.86	67.60	7.11
P53	60.63	67.17	64.76	62.09	61.99	62.92	61.99	61.27	62.56

PR AUC for High Dimensional Data Sets

Show Table

Data Set	AE	SOM	RBM	HiCS	LOF	FastABOD	LoOP	OC-SVM	KNN
Arrhythmia-2	29.37	27.94	1.66	1.64	3.83	3.98	3.51	4.90	3.04
Arrhythmia-5	30.61	36.15	5.00	4.14	30.62	29.53	29.85	30.56	22.37
Arrhythmia-10	48.54	48.24	10.31	13.84	45.36	42.49	45.55	40.55	33.24
Arrhythmia-20	51.97	50.62	20.35	15.78	49.12	48.14	48.90	49.92	46.98
Arrhythmia-46	75.32	74.54	44.98	44.67	74.36	72.83	71.63	73.68	71.50
InternetAds-2	1.58	32.78	1.91	94.52	34.86	32.79	3.51	4.90	3.04
InternetAds-5	5.12	45.86	5.21	97.67	52.56	40.25	51.12	33.54	32.87
InternetAds-10	7.69	47.89	10.12	41.13	51.93	35.74	46.09	29.40	35.11
InternetAds-19	14.29	54.67	18.91	1.44	55.24	41.86	44.48	29.46	33.64
ISOLET-2	44.45	66.93	29.51	4.47	70.95	29.54	51.78	25.65	42.91
ISOLET-4	49.97	69.76	33.59	9.08	74.21	24.07	47.62	47.61	23.96
ISOLET-8	59.60	78.61	40.50	14.36	43.81	21.71	31.45	46.55	28.59
ISOLET-14	64.48	75.40	53.84	29.06	40.03	33.68	27.15	47.53	36.62
ISOLET-20	64.26	63.55	40.50	36.92	34.56	37.72	25.59	48.72	41.45
MNIST	30.13	27.40	9.26	9.99	33.95	14.05	24.74	25.49	27.96
Musk	100.00	100.00	85.58	97.46	14.68	1.65	3.71	4.54	1.91
P53	1.04	1.29	1.20	1.25	1.06	1.13	1.06	1.06	1.30

Visualization

ROC AUC and PR AUC for Arrhythmia-2

Show Plot

ROC AUC and PR AUC for Arrhythmia-5

Show Plot

ROC AUC and PR AUC for Arrhythmia-10

Show Plot

ROC AUC and PR AUC for Arrhythmia-20

Show Plot

ROC AUC and PR AUC for Arrhythmia-46

Show Plot

ROC AUC and PR AUC for InternetAds-2

Show Plot

ROC AUC and PR AUC for InternetAds-5

Show Plot

ROC AUC and PR AUC for InternetAds-10

Show Plot

ROC AUC and PR AUC for InternetAds-19

Show Plot

ROC AUC and PR AUC for ISOLET-2

Show Plot

ROC AUC and PR AUC for ISOLET-4

Show Plot

ROC AUC and PR AUC for ISOLET-8

Show Plot

ROC AUC and PR AUC for ISOLET-14

Show Plot

ROC AUC and PR AUC for ISOLET-20

Show Plot

ROC AUC and PR AUC for MNIST

Show Plot

ROC AUC and PR AUC for Musk

Show Plot

ROC AUC and PR AUC for P53

Show Plot

Parameter Evaluation

In this section we present the complete tables of parameter evaluation values for Autoencoder, Restricted Boltzmann Machine, and Self-Organising Map.

Parameter Evaluation for Autoencoder (AE)

Show Table

Parameter	Epochs	$D_{Epsilon,n^e}$
0.4	10	0.
0.4	20	0.
0.4	100	0.
0.4	1,000	0.
0.5	10	0.0519
0.5	20	0.0405
0.5	100	0.0527
0.5	1,000	0.0820
0.6	10	0.0478
0.6	20	0.0411
0.6	100	0.0476
0.6	1,000	0.0844
0.7	10	0.0501
0.7	20	0.0412
0.7	100	0.0527
0.7	1,000	0.0788
0.8	10	0.0409
0.8	20	0.0403
0.8	100	0.0512
0.8	1,000	0.0827
0.9	10	0.0424
0.9	20	0.0412
0.9	100	0.0539
0.9	1,000	0.0828

Parameter Evaluation for Restricted Boltzmann Machine (RBM)

Show Table

Parameter	Parameter	Epochs	$D_{delta,gamma,n^e}$
0.1	0.1	10	0.2318
0.1	0.1	20	0.2067
0.1	0.1	100	0.1700
0.1	0.1	1,000	0.1507
0.1	0.2	10	0.1221
0.1	0.2	20	0.0985
0.1	0.2	100	0.0994
0.1	0.2	1,000	0.2349
0.1	0.3	10	0.2038
0.1	0.3	20	0.1925
0.1	0.3	100	0.1453
0.1	0.3	1,000	0.1192
0.1	0.4	10	0.0971
0.1	0.4	20	0.1142
0.1	0.4	100	0.2313
0.1	0.4	1,000	0.2086
0.1	0.5	10	0.1857
0.1	0.5	20	0.1641
0.1	0.5	100	0.1136
0.1	0.5	1,000	0.0903
0.1	0.6	10	0.0920
0.1	0.6	20	0.0928
0.1	0.6	100	0.2010
0.1	0.6	1,000	0.1900
0.1	0.7	10	0.1587
0.1	0.7	20	0.1375
0.1	0.7	100	0.1000
0.1	0.7	1,000	0.0740
0.1	0.8	10	0.0778
0.1	0.8	20	0.2408
0.1	0.8	100	0.1873
0.1	0.8	1,000	0.1576
0.1	0.9	10	0.1426
0.1	0.9	20	0.1121
0.1	0.9	100	0.0764
0.1	0.9	1,000	0.0756
0.2	0.1	10	0.2412
0.2	0.1	20	0.2058
0.2	0.1	100	0.1660
0.2	0.1	1,000	0.1460
0.2	0.2	10	0.1305
0.2	0.2	20	0.1202
0.2	0.2	100	0.0955
0.2	0.2	1,000	0.2266
0.2	0.3	10	0.1975
0.2	0.3	20	0.1832
0.2	0.3	100	0.1439
0.2	0.3	1,000	0.1142
0.2	0.4	10	0.1031
0.2	0.4	20	0.0913
0.2	0.4	100	0.2398
0.2	0.4	1,000	0.2001
0.2	0.5	10	0.1883
0.2	0.5	20	0.1628
0.2	0.5	100	0.1143
0.2	0.5	1,000	0.1070
0.2	0.6	10	0.0751
0.2	0.6	20	0.0881
0.2	0.6	100	0.2084
0.2	0.6	1,000	0.1852
0.2	0.7	10	0.1508
0.2	0.7	20	0.1400
0.2	0.7	100	0.0953
0.2	0.7	1,000	0.0764
0.2	0.8	10	0.0923
0.2	0.8	20	0.2278
0.2	0.8	100	0.1755
0.2	0.8	1,000	0.1612
0.2	0.9	10	0.1429
0.2	0.9	20	0.1281
0.2	0.9	100	0.0822
0.2	0.9	1,000	0.0638
0.3	0.1	10	0.2381
0.3	0.1	20	0.2174
0.3	0.1	100	0.1635
0.3	0.1	1,000	0.1473
0.3	0.2	10	0.1232
0.3	0.2	20	0.1111
0.3	0.2	100	0.0986
0.3	0.2	1,000	0.2351
0.3	0.3	10	0.2138
0.3	0.3	20	0.1846
0.3	0.3	100	0.1429
0.3	0.3	1,000	0.1249
0.3	0.4	10	0.0985
0.3	0.4	20	0.1104
0.3	0.4	100	0.2335
0.3	0.4	1,000	0.2191
0.3	0.5	10	0.1916
0.3	0.5	20	0.1622
0.3	0.5	100	0.1214
0.3	0.5	1,000	0.0978
0.3	0.6	10	0.0786
0.3	0.6	20	0.0657
0.3	0.6	100	0.2042
0.3	0.6	1,000	0.1806
0.3	0.7	10	0.1547
0.3	0.7	20	0.1338
0.3	0.7	100	0.0894
0.3	0.7	1,000	0.0781
0.3	0.8	10	0.0839
0.3	0.8	20	0.2311
0.3	0.8	100	0.1837
0.3	0.8	1,000	0.1461
0.3	0.9	10	0.1375
0.3	0.9	20	0.1237
0.3	0.9	100	0.0844
0.3	0.9	1,000	0.0736
0.4	0.1	10	0.2188
0.4	0.1	20	0.2109
0.4	0.1	100	0.1754
0.4	0.1	1,000	0.1455
0.4	0.2	10	0.1344
0.4	0.2	20	0.1146
0.4	0.2	100	0.0997
0.4	0.2	1,000	0.2355
0.4	0.3	10	0.2075
0.4	0.3	20	0.1836
0.4	0.3	100	0.1391
0.4	0.3	1,000	0.1193
0.4	0.4	10	0.1073
0.4	0.4	20	0.1049
0.4	0.4	100	0.2301
0.4	0.4	1,000	0.2083
0.4	0.5	10	0.1914
0.4	0.5	20	0.1555
0.4	0.5	100	0.1074
0.4	0.5	1,000	0.0930
0.4	0.6	10	0.0995
0.4	0.6	20	0.0852
0.4	0.6	100	0.2117
0.4	0.6	1,000	0.1779
0.4	0.7	10	0.1607
0.4	0.7	20	0.1287
0.4	0.7	100	0.0874
0.4	0.7	1,000	0.0886
0.4	0.8	10	0.0781
0.4	0.8	20	0.2340
0.4	0.8	100	0.1793
0.4	0.8	1,000	0.1592
0.4	0.9	10	0.1420
0.4	0.9	20	0.1240
0.4	0.9	100	0.1061
0.4	0.9	1,000	0.0759
0.5	0.1	10	0.2396
0.5	0.1	20	0.2065
0.5	0.1	100	0.1644
0.5	0.1	1,000	0.1414
0.5	0.2	10	0.1308
0.5	0.2	20	0.1139
0.5	0.2	100	0.0935
0.5	0.2	1,000	0.2268
0.5	0.3	10	0.2150
0.5	0.3	20	0.1950
0.5	0.3	100	0.1395
0.5	0.3	1,000	0.1150
0.5	0.4	10	0.1029
0.5	0.4	20	0.1019
0.5	0.4	100	0.2335
0.5	0.4	1,000	0.2097
0.5	0.5	10	0.1811
0.5	0.5	20	0.1583
0.5	0.5	100	0.1071
0.5	0.5	1,000	0.1072
0.5	0.6	10	0.0834
0.5	0.6	20	0.0865
0.5	0.6	100	0.2110
0.5	0.6	1,000	0.1754
0.5	0.7	10	0.1629
0.5	0.7	20	0.1280
0.5	0.7	100	0.0971
0.5	0.7	1,000	0.0797
0.5	0.8	10	0.0784
0.5	0.8	20	0.2378
0.5	0.8	100	0.1833
0.5	0.8	1,000	0.1388
0.5	0.9	10	0.1331
0.5	0.9	20	0.1081
0.5	0.9	100	0.0901
0.5	0.9	1,000	0.0675
0.6	0.1	10	0.2311
0.6	0.1	20	0.2155
0.6	0.1	100	0.1667
0.6	0.1	1,000	0.1556
0.6	0.2	10	0.1351
0.6	0.2	20	0.1135
0.6	0.2	100	0.1056
0.6	0.2	1,000	0.2303
0.6	0.3	10	0.2089
0.6	0.3	20	0.1915
0.6	0.3	100	0.1509
0.6	0.3	1,000	0.1183
0.6	0.4	10	0.1070
0.6	0.4	20	0.0856
0.6	0.4	100	0.2253
0.6	0.4	1,000	0.2090
0.6	0.5	10	0.1864
0.6	0.5	20	0.1631
0.6	0.5	100	0.1118
0.6	0.5	1,000	0.0971
0.6	0.6	10	0.0778
0.6	0.6	20	0.0817
0.6	0.6	100	0.2173
0.6	0.6	1,000	0.1775
0.6	0.7	10	0.1501
0.6	0.7	20	0.1241
0.6	0.7	100	0.0870
0.6	0.7	1,000	0.0797
0.6	0.8	10	0.0847
0.6	0.8	20	0.2358
0.6	0.8	100	0.1839
0.6	0.8	1,000	0.1587
0.6	0.9	10	0.1371
0.6	0.9	20	0.1132
0.6	0.9	100	0.0997
0.6	0.9	1,000	0.0863
0.7	0.1	10	0.2339
0.7	0.1	20	0.2093
0.7	0.1	100	0.1595
0.7	0.1	1,000	0.1483
0.7	0.2	10	0.1348
0.7	0.2	20	0.1075
0.7	0.2	100	0.0890
0.7	0.2	1,000	0.2187
0.7	0.3	10	0.2239
0.7	0.3	20	0.1798
0.7	0.3	100	0.1406
0.7	0.3	1,000	0.1169
0.7	0.4	10	0.0912
0.7	0.4	20	0.0944
0.7	0.4	100	0.2441
0.7	0.4	1,000	0.2151
0.7	0.5	10	0.1833
0.7	0.5	20	0.1663
0.7	0.5	100	0.1179
0.7	0.5	1,000	0.0972
0.7	0.6	10	0.0904
0.7	0.6	20	0.0869
0.7	0.6	100	0.2039
0.7	0.6	1,000	0.1943
0.7	0.7	10	0.1590
0.7	0.7	20	0.1289
0.7	0.7	100	0.0913
0.7	0.7	1,000	0.0811
0.7	0.8	10	0.0893
0.7	0.8	20	0.2375
0.7	0.8	100	0.1887
0.7	0.8	1,000	0.1484
0.7	0.9	10	0.1430
0.7	0.9	20	0.1059
0.7	0.9	100	0.0782
0.7	0.9	1,000	0.0755
0.8	0.1	10	0.2353
0.8	0.1	20	0.2178
0.8	0.1	100	0.1627
0.8	0.1	1,000	0.1498
0.8	0.2	10	0.1203
0.8	0.2	20	0.1233
0.8	0.2	100	0.0925
0.8	0.2	1,000	0.2357
0.8	0.3	10	0.2109
0.8	0.3	20	0.1863
0.8	0.3	100	0.1432
0.8	0.3	1,000	0.1246
0.8	0.4	10	0.1082
0.8	0.4	20	0.1023
0.8	0.4	100	0.2293
0.8	0.4	1,000	0.2125
0.8	0.5	10	0.1751
0.8	0.5	20	0.1630
0.8	0.5	100	0.1205
0.8	0.5	1,000	0.0936
0.8	0.6	10	0.0884
0.8	0.6	20	0.0804
0.8	0.6	100	0.2186
0.8	0.6	1,000	0.1776
0.8	0.7	10	0.1602
0.8	0.7	20	0.1250
0.8	0.7	100	0.0953
0.8	0.7	1,000	0.0816
0.8	0.8	10	0.0778
0.8	0.8	20	0.2266
0.8	0.8	100	0.1781
0.8	0.8	1,000	0.1522
0.8	0.9	10	0.1370
0.8	0.9	20	0.1122
0.8	0.9	100	0.0597
0.8	0.9	1,000	0.0732
0.9	0.1	10	0.2355
0.9	0.1	20	0.2089
0.9	0.1	100	0.1710
0.9	0.1	1,000	0.1554
0.9	0.2	10	0.1355
0.9	0.2	20	0.1155
0.9	0.2	100	0.1068
0.9	0.2	1,000	0.2370
0.9	0.3	10	0.2120
0.9	0.3	20	0.1828
0.9	0.3	100	0.1432
0.9	0.3	1,000	0.1210
0.9	0.4	10	0.1080
0.9	0.4	20	0.0945
0.9	0.4	100	0.2296
0.9	0.4	1,000	0.2073
0.9	0.5	10	0.1874
0.9	0.5	20	0.1612
0.9	0.5	100	0.1059
0.9	0.5	1,000	0.0900
0.9	0.6	10	0.0940
0.9	0.6	20	0.0753
0.9	0.6	100	0.2092
0.9	0.6	1,000	0.1892
0.9	0.7	10	0.1628
0.9	0.7	20	0.1460
0.9	0.7	100	0.0971
0.9	0.7	1,000	0.0915
0.9	0.8	10	0.0786
0.9	0.8	20	0.2337
0.9	0.8	100	0.1866
0.9	0.8	1,000	0.1683
0.9	0.9	10	0.1431
0.9	0.9	20	0.1117
0.9	0.9	100	0.0765
0.9	0.9	1,000	0.0800

Parameter Evaluation for Self-Organising Map (SOM)

Show Table

Parameter	Epochs	$D_{n,n^e}$
1	10	0.0496
1	20	0.0649
1	100	0.0496
1	1,000	0.0496
2	10	0.0221
2	20	0.0330
2	100	0.0223
2	1,000	0.0223
3	10	0.0315
3	20	0.0466
3	100	0.0332
3	1,000	0.0330
4	10	0.0399
4	20	0.0509
4	100	0.0543
4	1,000	0.0554
5	10	0.0768
5	20	0.0932
5	100	0.0850
5	1,000	0.0859
6	10	0.0866
6	20	0.0906
6	100	0.0901
6	1,000	0.0890
7	10	0.1040
7	20	0.1186
7	100	0.1117
7	1,000	0.1132
8	10	0.0995
8	20	0.1103
8	100	0.1107
8	1,000	0.1073
9	10	0.1144
9	20	0.1207
9	100	0.1095
9	1,000	0.1103
10	10	0.1314
10	20	0.1425
10	100	0.1166
10	1,000	0.1144
11	10	0.1807
11	20	0.1724
11	100	0.1596
11	1,000	0.1642
12	10	0.1815
12	20	0.1771
12	100	0.1742
12	1,000	0.1667
13	10	0.1935
13	20	0.1919
13	100	0.1784
13	1,000	0.1826
14	10	0.1967
14	20	0.1898
14	100	0.1885
14	1,000	0.1969
15	10	0.2003
15	20	0.1965
15	100	0.2023
15	1,000	0.2083
16	10	0.2192
16	20	0.2230
16	100	0.2111
16	1,000	0.2175
17	10	0.2241
17	20	0.2218
17	100	0.2176
17	1,000	0.2075
18	10	0.2269
18	20	0.2271
18	100	0.2136
18	1,000	0.2234
19	10	0.2483
19	20	0.2331
19	100	0.2493
19	1,000	0.2291
20	10	0.2339
20	20	0.2227
20	100	0.2348
20	1,000	0.2346

Acknowledgements

This repository features code and data sets from other projects. We would like to acknowledge the following contributions:

Code

ELKI Data Mining Toolkit by Erich Schubert, Alexander Koos, Tobias Emrich, Andreas Züﬂe, Klaus Arthur Schmid, Arthur Zimek. A Framework for Clustering Uncertain Data. Proceedings of the VLDB Endowment, 8(12): pp.1976–1979 (2015). VLDB.
Somoclu by Peter Wittek, Shi Chao Gao, Ik Soo Lim, Li Zhao. Somoclu: An Efficient Parallel Library for Self-Organizing Maps. Journal of Statistical Software, 78(9), pp.1-21 (2017). DOI:10.18637/jss.v078.i09. arXiv:1305.1422.

Data Sets

DAMI by Guilherme O. Campos, Arthur Zimek, Jörg Sander, Ricardo J. G. B. Campello, Barbora Micenková, Erich Schubert, Ira Assent, Michael E. Houle (2016). On the Evaluation of Unsupervised Outlier Detection: Measures, Datasets, and an Empirical Study. Data Mining and Knowledge Discovery 30(4), Springer Nature, pp.891-927 (2016). DOI:10.1007/s10618-015-0444-8
ODDS Library by Shebuti Rayana. Stony Brook University, Department of Computer Sciences, Stony Brook, NY (2016).
UCI Machine Learning Repository by Dheeru Dua, Casey Graff. University of California, Irvine, School of Information and Computer Sciences (2017).

Name		Name	Last commit message	Last commit date
datasets		datasets
images		images
nnohd		nnohd
LICENSE		LICENSE
README.md		README.md
requirements.txt		requirements.txt

Latest commit

History

Folders and files

Repository files navigation

Unsupervised Artificial Neural Networks for Outlier Detection in High-Dimensional Data

Supplementary Material

Quick Start

Prerequisites

Run the Outlier Detector

Autoencoder

Restricted Boltzmann Machine

Self-Organising Map

All Methods

Data Set Structure

ARFF Files

CSV Files

MATLAB Files

Implementation Release

Project Structure

Root Folder

Application Structure

Dependencies

Additional Material

Evaluated Data Sets

Dataset Evaluation

ROC AUC for High Dimensional Data Sets

PR AUC for High Dimensional Data Sets

Visualization

ROC AUC and PR AUC for Arrhythmia-2

ROC AUC and PR AUC for Arrhythmia-5

ROC AUC and PR AUC for Arrhythmia-10

ROC AUC and PR AUC for Arrhythmia-20

ROC AUC and PR AUC for Arrhythmia-46

ROC AUC and PR AUC for InternetAds-2

ROC AUC and PR AUC for InternetAds-5

ROC AUC and PR AUC for InternetAds-10

ROC AUC and PR AUC for InternetAds-19

ROC AUC and PR AUC for ISOLET-2

ROC AUC and PR AUC for ISOLET-4

ROC AUC and PR AUC for ISOLET-8

ROC AUC and PR AUC for ISOLET-14

ROC AUC and PR AUC for ISOLET-20

ROC AUC and PR AUC for MNIST

ROC AUC and PR AUC for Musk

ROC AUC and PR AUC for P53

Parameter Evaluation

Parameter Evaluation for Autoencoder (AE)

Parameter Evaluation for Restricted Boltzmann Machine (RBM)

Parameter Evaluation for Self-Organising Map (SOM)

Acknowledgements

Code

Data Sets

About

Resources

Stars

Watchers

Forks

Releases

Packages

Used by

Contributors

Languages