Here we present a dataset, MNIST4OD, of large size (number of dimensions and number of instances) suitable for Outliers Detection task.The dataset is based on the famous MNIST dataset (http://yann.lecun.com/exdb/mnist/).We build MNIST4OD in the following way:To distinguish between outliers and inliers, we choose the images belonging to a digit as inliers (e.g. digit 1) and we sample with uniform probability on the remaining images as outliers such as their number is equal to 10% of that of inliers. We repeat this dataset generation process for all digits. For implementation simplicity we then flatten the images (28 X 28) into vectors.Each file MNIST_x.csv.gz contains the corresponding dataset where the inlier class is equal to x.The data contains one instance (vector) in each line where the last column represents the outlier label (yes/no) of the data point. The data contains also a column which indicates the original image class (0-9).See the following numbers for a complete list of the statistics of each datasets ( Name | Instances | Dimensions | Number of Outliers in % ):MNIST_0 | 7594 | 784 | 10MNIST_1 | 8665 | 784 | 10MNIST_2 | 7689 | 784 | 10MNIST_3 | 7856 | 784 | 10MNIST_4 | 7507 | 784 | 10MNIST_5 | 6945 | 784 | 10MNIST_6 | 7564 | 784 | 10MNIST_7 | 8023 | 784 | 10MNIST_8 | 7508 | 784 | 10MNIST_9 | 7654 | 784 | 10

There has been a tremendous increase in the volume of sensor data collected over the last decade for different monitoring tasks. For example, petabytes of earth science data are collected from modern satellites, in-situ sensors and different climate models. Similarly, huge amount of flight operational data is downloaded for different commercial airlines. These different types of datasets need to be analyzed for finding outliers. Information extraction from such rich data sources using advanced data mining methodologies is a challenging task not only due to the massive volume of data, but also because these datasets are physically stored at different geographical locations with only a subset of features available at any location. Moving these petabytes of data to a single location may waste a lot of bandwidth. To solve this problem, in this paper, we present a novel algorithm which can identify outliers in the entire data without moving all the data to a single location. The method we propose only centralizes a very small sample from the different data subsets at different locations. We analytically prove and experimentally verify that the algorithm offers high accuracy compared to complete centralization with only a fraction of the communication cost. We show that our algorithm is highly relevant to both earth sciences and aeronautics by describing applications in these domains. The performance of the algorithm is demonstrated on two large publicly available datasets: (1) the NASA MODIS satellite images and (2) a simulated aviation dataset generated by the ‘Commercial Modular Aero-Propulsion System Simulation’ (CMAPSS).

Identification of errors or anomalous values, collectively considered outliers, assists in exploring data or through removing outliers improves statistical analysis. In biomechanics, outlier detection methods have explored the ‘shape’ of the entire cycles, although exploring fewer points using a ‘moving-window’ may be advantageous. Hence, the aim was to develop a moving-window method for detecting trials with outliers in intra-participant time-series data. Outliers were detected through two stages for the strides (mean 38 cycles) from treadmill running. Cycles were removed in stage 1 for one-dimensional (spatial) outliers at each time point using the median absolute deviation, and in stage 2 for two-dimensional (spatial–temporal) outliers using a moving window standard deviation. Significance levels of the t-statistic were used for scaling. Fewer cycles were removed with smaller scaling and smaller window size, requiring more stringent scaling at stage 1 (mean 3.5 cycles removed for 0.0001 scaling) than at stage 2 (mean 2.6 cycles removed for 0.01 scaling with a window size of 1). Settings in the supplied Matlab code should be customised to each data set, and outliers assessed to justify whether to retain or remove those cycles. The method is effective in identifying trials with outliers in intra-participant time series data.

Dataset Card for "mnist-outlier"

📚 This dataset is an enriched version of the MNIST Dataset. The workflow is described in the medium article: Changes of Embeddings during Fine-Tuning of Transformers.

  Explore the Dataset

The open source data curation tool Renumics Spotlight allows you to explorer this dataset. You can find a Hugging Face Space running Spotlight with this dataset here: https://huggingface.co/spaces/renumics/mnist-outlier.

Or you can explorer it locally:… See the full description on the dataset page: https://huggingface.co/datasets/renumics/mnist-outlier.

Cylindrical data are bivariate data formed from the combination of circular and linear variables. Identifying outliers is a crucial step in any data analysis work. This paper proposes a new distribution-free procedure to detect outliers in cylindrical data using the Mahalanobis distance concept. The use of Mahalanobis distance incorporates the correlation between the components of the cylindrical distribution, which had not been accounted for in the earlier papers on outlier detection in cylindrical data. The threshold for declaring an observation to be an outlier can be obtained via parametric or non-parametric bootstrap, depending on whether the underlying distribution is known or unknown. The performance of the proposed method is examined via extensive simulations from the Johnson-Wehrly distribution. The proposed method is applied to two real datasets, and the outliers are identified in those datasets.

These files are supplements to the paper titled 'A Robust Two-step Method for Detection of Outlier Sets'.This paper identifies and addresses the need for a robust method that identifies sets of points that collectively deviate from typical patterns in a dataset, which it calls "outlier sets'', while excluding individual points from detection. This new methodology, Outlier Set Two-step Identification (OSTI) employs a two-step approach to detect and label these outlier sets. First, it uses Gaussian Mixture Models for probabilistic clustering, identifying candidate outlier sets based on cluster weights below a predetermined threshold. Second, OSTI measures the Inter-cluster Mahalanobis distance between each candidate outlier set's centroid and the overall dataset mean. OSTI then tests the null hypothesis that this distance does not significantly differ from its theoretical chi-square distribution, enabling the formal detection of outlier sets. We test OSTI systematically on 8,000 synthetic 2D datasets across various inlier configurations and thousands of possible outlier set characteristics. Results show OSTI robustly and consistently detects outlier sets with an average F1 score of 0.92 and an average purity (the degree to which outlier sets identified correspond to those generated synthetically, i.e., our ground truth) of 98.58%. We also compare OSTI with state-of-the-art outlier detection methods, to illuminate how OSTI fills a gap as a tool for the exclusive detection of outlier sets.

We present a set of novel algorithms which we call sequenceMiner, that detect and characterize anomalies in large sets of high-dimensional symbol sequences that arise from recordings of switch sensors in the cockpits of commercial airliners. While the algorithms we present are general and domain-independent, we focus on a specific problem that is critical to determining system-wide health of a fleet of aircraft. The approach taken uses unsupervised clustering of sequences using the normalized length of he longest common subsequence (nLCS) as a similarity measure, followed by a detailed analysis of outliers to detect anomalies. In this method, an outlier sequence is defined as a sequence that is far away from a cluster. We present new algorithms for outlier analysis that provide comprehensible indicators as to why a particular sequence is deemed to be an outlier. The algorithm provides a coherent description to an analyst of the anomalies in the sequence when compared to more normal sequences. The final section of the paper demonstrates the effectiveness of sequenceMiner for anomaly detection on a real set of discrete sequence data from a fleet of commercial airliners. We show that sequenceMiner discovers actionable and operationally significant safety events. We also compare our innovations with standard HiddenMarkov Models, and show that our methods are superior

Dataset Card for "cifar10-outlier"

📚 This dataset is an enriched version of the CIFAR-10 Dataset. The workflow is described in the medium article: Changes of Embeddings during Fine-Tuning of Transformers.

  Explore the Dataset

The open source data curation tool Renumics Spotlight allows you to explorer this dataset. You can find a Hugging Face Spaces running Spotlight with this dataset here:

Full Version (High hardware requirement)… See the full description on the dataset page: https://huggingface.co/datasets/renumics/cifar10-outlier.

The "Amsterdam Library of Object Images" is a collection of 110250 images of 1000 small objects, taken under various light conditions and rotation angles. All objects were placed on a black background. Thus the images are taken under rather uniform conditions, which means there is little uncontrolled bias in the data set (unless mixed with other sources). They do however not resemble a "typical" image collection. The data set has a rather unique property for its size: there are around 100 different images of each object, so it is well suited for clustering. By downsampling some objects it can also be used for outlier detection. For multi-view research, we offer a number of different feature vector sets for evaluating this data set.

The Multi-Domain Outlier Detection Dataset contains datasets for conducting outlier detection experiments for four different application domains:

Astrophysics - detecting anomalous observations in the Dark Energy Survey (DES) catalog (data type: feature vectors)

Planetary science - selecting novel geologic targets for follow-up observation onboard the Mars Science Laboratory (MSL) rover (data type: grayscale images)

Earth science: detecting anomalous samples in satellite time series corresponding to ground-truth observations of maize crops (data type: time series/feature vectors)

Fashion-MNIST/MNIST: benchmark task to detect anomalous MNIST images among Fashion-MNIST images (data type: grayscale images)

Each dataset contains a "fit" dataset (used for fitting or training outlier detection models), a "score" dataset (used for scoring samples used to evaluate model performance, analogous to test set), and a label dataset (indicates whether samples in the score dataset are considered outliers or not in the domain of each dataset).

To read more about the datasets and how they are used for outlier detection, or to cite this dataset in your own work, please see the following citation:

Kerner, H. R., Rebbapragada, U., Wagstaff, K. L., Lu, S., Dubayah, B., Huff, E., Lee, J., Raman, V., and Kulshrestha, S. (2022). Domain-agnostic Outlier Ranking Algorithms (DORA)-A Configurable Pipeline for Facilitating Outlier Detection in Scientific Datasets. Under review for Frontiers in Astronomy and Space Sciences.

OSU_SnowCourse Summary: Manual snow course observations were collected over WY 2012-2014 from four paired forest-open sites chosen to span a broad elevation range. Study sites were located in the upper McKenzie (McK) River watershed, approximately 100 km east of Corvallis, Oregon, on the western slope of the Cascade Range and in the Middle Fork Willamette (MFW) watershed, located to the south of the McKenzie. The sites were designated based on elevation, with a range of 1110-1480 m. Distributed snow depth and snow water equivalent (SWE) observations were collected via monthly manual snow courses from 1 November through 1 April and bi-weekly thereafter. Snow courses spanned 500 m of forested terrain and 500 m of adjacent open terrain. Snow depth observations were collected approximately every 10 m and SWE was measured every 100 m along the snow courses with a federal snow sampler. These data are raw observations and have not been quality controlled in any way. Distance along the transect was estimated in the field. OSU_SnowDepth Summary: 10-minute snow depth observations collected at OSU met stations in the upper McKenzie River Watershed and the Middle Fork Willamette Watershed during Water Years 2012-2014. Each meterological tower was deployed to represent either a forested or an open area at a particular site, and generally the locations were paired, with a meterological station deployed in the forest and in the open area at a single site. These data were collected in conjunction with manual snow course observations, and the meterological stations were located in the approximate center of each forest or open snow course transect. These data have undergone basic quality control. See manufacturer specifications for individual instruments to determine sensor accuracy. This file was compiled from individual raw data files (named "RawData.txt" within each site and year directory) provided by OSU, along with metadata of site attributes. We converted the Excel-based timestamp (seconds since origin) to a date, changed the NaN flags for missing data to NA, and added site attributes such as site name and cover. We replaced positive values with NA, since snow depth values in raw data are negative (i.e., flipped, with some correction to use the height of the sensor as zero). Thus, positive snow depth values in the raw data equal negative snow depth values. Second, the sign of the data was switched to make them positive. Then, the smooth.m (MATLAB) function was used to roughly smooth the data, with a moving window of 50 points. Third, outliers were removed. All values higher than the smoothed values +10, were replaced with NA. In some cases, further single point outliers were removed. OSU_Met Summary: Raw, 10-minute meteorological observations collected at OSU met stations in the upper McKenzie River Watershed and the Middle Fork Willamette Watershed during Water Years 2012-2014. Each meterological tower was deployed to represent either a forested or an open area at a particular site, and generally the locations were paired, with a meterological station deployed in the forest and in the open area at a single site. These data were collected in conjunction with manual snow course observations, and the meteorological stations were located in the approximate center of each forest or open snow course transect. These stations were deployed to collect numerous meteorological variables, of which snow depth and wind speed are included here. These data are raw datalogger output and have not been quality controlled in any way. See manufacturer specifications for individual instruments to determine sensor accuracy. This file was compiled from individual raw data files (named "RawData.txt" within each site and year directory) provided by OSU, along with metadata of site attributes. We converted the Excel-based timestamp (seconds since origin) to a date, changed the NaN and 7999 flags for missing data to NA, and added site attributes such as site name and cover. OSU_Location Summary: Location Metadata for manual snow course observations and meteorological sensors. These data are compiled from GPS data for which the horizontal accuracy is unknown, and from processed hemispherical photographs. They have not been quality controlled in any way.

Abstract: Building health management is an important part in running an efficient and cost-effective building. Many problems in a building’s system can go undetected for long periods of time, leading to expensive repairs or wasted resources. This project aims to help detect and diagnose the building‘s health with data driven methods throughout the day. Orca and IMS are two state of the art algorithms that observe an array of building health sensors and provide feedback on the overall system’s health as well as localize the problem to one, or possibly two, components. With this level of feedback the hope is to quickly identify problems and provide appropriate maintenance while reducing the number of complaints and service calls. Introduction: To prepare these technologies for the new installation, the proposed methods are being tested on a current system that behaves similarly to the future green building. Building 241 was determined to best resemble the proposed building 232 and therefore was chosen for this study. Building 241 is currently outfitted with 34 sensors that monitor the heating & cooling temperatures for the air and water systems as well as other various subsystem states. The daily sensor recordings were logged and sent to the IDU group for analysis. The period of analysis was focused from July 1st through August 10th 2009. Methodology: The two algorithms used for analysis were Orca and IMS. Both methods look for anomalies using a distanced based scoring approach. Orca has the ability to use a single data set and find outliers within that data set. This tactic was applied to each day. After scoring each time sample throughout a given day the Orca score profiles were compared by computing the correlation against all other days. Days with high overall correlations were considered normal however days with lower overall correlations were more anomalous. IMS, on the other hand, needs a normal set of data to build a model, which can be applied to a set of test data to asses how anomaly the particular data set is. The typical days identified by Orca were used as the reference/training set for IMS, while all the other days were passed through IMS resulting in an anomaly score profile for each day. The mean of the IMS score profile was then calculated for each day to produce a summary IMS score. These summary scores were ranked and the top outliers were identified (see Figure 1). Once the anomalies were identified the contributing parameters were then ranked by the algorithm. Analysis: The contributing parameters identified by IMS were localized to the return air temperature duct system. -7/03/09 (Figure 2 & 3) AHU-1 Return Air Temperature (RAT) Calculated Average Return Air Temperature -7/19/09 (Figure 3 & 4) AHU-2 Return Air Temperature (RAT) Calculated Average Return Air Temperature IMS identified significantly higher temperatures compared to other days during the month of July and August. Conclusion: The proposed algorithms Orca and IMS have shown that they were able to pick up significant anomalies in the building system as well as diagnose the anomaly by identifying the sensor values that were anomalous. In the future these methods can be used on live streaming data and produce a real time anomaly score to help building maintenance with detection and diagnosis of problems.

The Controlled Anomalies Time Series (CATS) Dataset consists of commands, external stimuli, and telemetry readings of a simulated complex dynamical system with 200 injected anomalies.

The CATS Dataset exhibits a set of desirable properties that make it very suitable for benchmarking Anomaly Detection Algorithms in Multivariate Time Series [1]:

• Multivariate (17 variables) including sensors reading and control signals. It simulates the operational behaviour of an arbitrary complex system including:
  • 4 Deliberate Actuations / Control Commands sent by a simulated operator / controller, for instance, commands of an operator to turn ON/OFF some equipment.
  • 3 Environmental Stimuli / External Forces acting on the system and affecting its behaviour, for instance, the wind affecting the orientation of a large ground antenna.
  • 10 Telemetry Readings representing the observable states of the complex system by means of sensors, for instance, a position, a temperature, a pressure, a voltage, current, humidity, velocity, acceleration, etc.
• 5 million timestamps. Sensors readings are at 1Hz sampling frequency.
  • 1 million nominal observations (the first 1 million datapoints). This is suitable to start learning the "normal" behaviour.
  • 4 million observations that include both nominal and anomalous segments. This is suitable to evaluate both semi-supervised approaches (novelty detection) as well as unsupervised approaches (outlier detection).
• 200 anomalous segments. One anomalous segment may contain several successive anomalous observations / timestamps. Only the last 4 million observations contain anomalous segments.
• Different types of anomalies to understand what anomaly types can be detected by different approaches. The categories are available in the dataset and in the metadata.
• Fine control over ground truth. As this is a simulated system with deliberate anomaly injection, the start and end time of the anomalous behaviour is known very precisely. In contrast to real world datasets, there is no risk that the ground truth contains mislabelled segments which is often the case for real data.
• Suitable for root cause analysis. In addition to the anomaly category, the time series channel in which the anomaly first developed itself is recorded and made available as part of the metadata. This can be useful to evaluate the performance of algorithm to trace back anomalies to the right root cause channel.
• Affected channels. In addition to the knowledge of the root cause channel in which the anomaly first developed itself, we provide information of channels possibly affected by the anomaly. This can also be useful to evaluate the explainability of anomaly detection systems which may point out to the anomalous channels (root cause and affected).
• Obvious anomalies. The simulated anomalies have been designed to be "easy" to be detected for human eyes (i.e., there are very large spikes or oscillations), hence also detectable for most algorithms. It makes this synthetic dataset useful for screening tasks (i.e., to eliminate algorithms that are not capable to detect those obvious anomalies). However, during our initial experiments, the dataset turned out to be challenging enough even for state-of-the-art anomaly detection approaches, making it suitable also for regular benchmark studies.
• Context provided. Some variables can only be considered anomalous in relation to other behaviours. A typical example consists of a light and switch pair. The light being either on or off is nominal, the same goes for the switch, but having the switch on and the light off shall be considered anomalous. In the CATS dataset, users can choose (or not) to use the available context, and external stimuli, to test the usefulness of the context for detecting anomalies in this simulation.
• Pure signal ideal for robustness-to-noise analysis. The simulated signals are provided without noise: while this may seem unrealistic at first, it is an advantage since users of the dataset can decide to add on top of the provided series any type of noise and choose an amplitude. This makes it well suited to test how sensitive and robust detection algorithms are against various levels of noise.
• No missing data. You can drop whatever data you want to assess the impact of missing values on your detector with respect to a clean baseline.

Change Log

Version 2

• Metadata: we include a metadata.csv with information about:
  • Anomaly categories
  • Root cause channel (signal in which the anomaly is first visible)
  • Affected channel (signal in which the anomaly might propagate) through coupled system dynamics
• Removal of anomaly overlaps: version 1 contained anomalies which overlapped with each other resulting in only 190 distinct anomalous segments. Now, there are no more anomaly overlaps.
• Two data files: CSV and parquet for convenience.

[1] Example Benchmark of Anomaly Detection in Time Series: “Sebastian Schmidl, Phillip Wenig, and Thorsten Papenbrock. Anomaly Detection in Time Series: A Comprehensive Evaluation. PVLDB, 15(9): 1779 - 1797, 2022. doi:10.14778/3538598.3538602”

About Solenix

Solenix is an international company providing software engineering, consulting services and software products for the space market. Solenix is a dynamic company that brings innovative technologies and concepts to the aerospace market, keeping up to date with technical advancements and actively promoting spin-in and spin-out technology activities. We combine modern solutions which complement conventional practices. We aspire to achieve maximum customer satisfaction by fostering collaboration, constructivism, and flexibility.

Usage notes

Abstract: Building health management is an important part in running an efficient and cost-effective building. Many problems in a building’s system can go undetected for long periods of time, leading to expensive repairs or wasted resources. This project aims to help detect and diagnose the building‘s health with data driven methods throughout the day. Orca and IMS are two state of the art algorithms that observe an array of building health sensors and provide feedback on the overall system’s health as well as localize the problem to one, or possibly two, components. With this level of feedback the hope is to quickly identify problems and provide appropriate maintenance while reducing the number of complaints and service calls. Introduction: To prepare these technologies for the new installation, the proposed methods are being tested on a current system that behaves similarly to the future green building. Building 241 was determined to best resemble the proposed building 232 and therefore was chosen for this study. Building 241 is currently outfitted with 34 sensors that monitor the heating & cooling temperatures for the air and water systems as well as other various subsystem states. The daily sensor recordings were logged and sent to the IDU group for analysis. The period of analysis was focused from July 1st through August 10th 2009. Methodology: The two algorithms used for analysis were Orca and IMS. Both methods look for anomalies using a distanced based scoring approach. Orca has the ability to use a single data set and find outliers within that data set. This tactic was applied to each day. After scoring each time sample throughout a given day the Orca score profiles were compared by computing the correlation against all other days. Days with high overall correlations were considered normal however days with lower overall correlations were more anomalous. IMS, on the other hand, needs a normal set of data to build a model, which can be applied to a set of test data to asses how anomaly the particular data set is. The typical days identified by Orca were used as the reference/training set for IMS, while all the other days were passed through IMS resulting in an anomaly score profile for each day. The mean of the IMS score profile was then calculated for each day to produce a summary IMS score. These summary scores were ranked and the top outliers were identified (see Figure 1). Once the anomalies were identified the contributing parameters were then ranked by the algorithm. Analysis: The contributing parameters identified by IMS were localized to the return air temperature duct system. -7/03/09 (Figure 2 & 3) AHU-1 Return Air Temperature (RAT) Calculated Average Return Air Temperature -7/19/09 (Figure 3 & 4) AHU-2 Return Air Temperature (RAT) Calculated Average Return Air Temperature IMS identified significantly higher temperatures compared to other days during the month of July and August. Conclusion: The proposed algorithms Orca and IMS have shown that they were able to pick up significant anomalies in the building system as well as diagnose the anomaly by identifying the sensor values that were anomalous. In the future these methods can be used on live streaming data and produce a real time anomaly score to help building maintenance with detection and diagnosis of problems.

Our primary goal is to automatically analyze textual reports from the Aviation Safety Reporting System (ASRS) database to detect/discover the anomaly categories reported by the pilots, and to assign each report to the appropriate category/categories. We have used two state-of-the-art models for text analysis: (i) mixture of von Mises Fisher (movMF) distributions, and (ii) latent Dirichlet allocation (LDA) on a subset of all ASRS reports. The models achieve a reasonably high performance in discovering anomaly categories and clustering reports. Each category is represented by the most representative words with the highest probability in this category. In addition, since the inference algorithm for LDA was somewhat slow, we have developed a new fast LDA algorithm which is 5-10 times more efficient than the original one, therefore more applicable for the practical use. Further, we have developed a simple visualization tool based on non-linear manifold embedding (ISOMAP) to generate a 2-d visual representation of each report based on its content/topics, which gives a direct view of the structure of the whole dataset as well as the outliers.

R code used for each data set to perform negative binomial regression, calculate overdispersion statistic, generate summary statistics, remove outliers

1. NCCTG Lung cancer datasetSurvival in patients with advanced lung cancer from the North Central Cancer Treatment Group. Performance scores rate how well the patient can perform usual daily activities.2.CNV measurements of CNV of GBM This dataset records the information about copy number variation of Glioblastoma (GBM).Abstract:In biology and medicine, conservative patient and data collection malpractice can lead to missing or incorrect values in patient registries, which can affect both diagnosis and prognosis. Insufficient or biased patient information significantly impedes the sensitivity and accuracy of predicting cancer survival. In bioinformatics, making a best guess of the missing values and identifying the incorrect values are collectively called “imputation”. Existing imputation methods work by establishing a model based on the data mechanism of the missing values. Existing imputation methods work well under two assumptions: 1) the data is missing completely at random, and 2) the percentage of missing values is not high. These are not cases found in biomedical datasets, such as the Cancer Genome Atlas Glioblastoma Copy-Number Variant dataset (TCGA: 108 columns), or the North Central Cancer Treatment Group Lung Cancer (NCCTG) dataset (NCCTG: 9 columns). We tested six existing imputation methods, but only two of them worked with these datasets: The Last Observation Carried Forward (LOCF) and K-nearest Algorithm (KNN). Predictive Mean Matching (PMM) and Classification and Regression Trees (CART) worked only with the NCCTG lung cancer dataset with fewer columns, except when the dataset contains 45% missing data. The quality of the imputed values using existing methods is bad because they do not meet the two assumptions.In our study, we propose a Restricted Boltzmann Machine (RBM)-based imputation method to cope with low randomness and the high percentage of the missing values. RBM is an undirected, probabilistic and parameterized two-layer neural network model, which is often used for extracting abstract information from data, especially for high-dimensional data with unknown or non-standard distributions. In our benchmarks, we applied our method to two cancer datasets: 1) NCCTG, and 2) TCGA. The running time, root mean squared error (RMSE) of the different methods were gauged. The benchmarks for the NCCTG dataset show that our method performs better than other methods when there is 5% missing data in the dataset, with 4.64 RMSE lower than the best KNN. For the TCGA dataset, our method achieved 0.78 RMSE lower than the best KNN.In addition to imputation, RBM can achieve simultaneous predictions. We compared the RBM model with four traditional prediction methods. The running time and area under the curve (AUC) were measured to evaluate the performance. Our RBM-based approach outperformed traditional methods. Specifically, the AUC was up to 19.8% higher than the multivariate logistic regression model in the NCCTG lung cancer dataset, and the AUC was higher than the Cox proportional hazard regression model, with 28.1% in the TCGA dataset.Apart from imputation and prediction, RBM models can detect outliers in one pass by allowing the reconstruction of all the inputs in the visible layer with in a single backward pass. Our results show that RBM models have achieved higher precision and recall on detecting outliers than other methods.

The Controlled Anomalies Time Series (CATS) Dataset consists of commands, external stimuli, and telemetry readings of a simulated complex dynamical system with 200 injected anomalies.

The CATS Dataset exhibits a set of desirable properties that make it very suitable for benchmarking Anomaly Detection Algorithms in Multivariate Time Series [1]:

• Multivariate (17 variables) including sensors reading and control signals. It simulates the operational behaviour of an arbitrary complex system including:
  • 4 Deliberate Actuations / Control Commands sent by a simulated operator / controller, for instance, commands of an operator to turn ON/OFF some equipment.
  • 3 Environmental Stimuli / External Forces acting on the system and affecting its behaviour, for instance, the wind affecting the orientation of a large ground antenna.
  • 10 Telemetry Readings representing the observable states of the complex system by means of sensors, for instance, a position, a temperature, a pressure, a voltage, current, humidity, velocity, acceleration, etc.
• 5 million timestamps. Sensors readings are at 1Hz sampling frequency.
  • 1 million nominal observations (the first 1 million datapoints). This is suitable to start learning the "normal" behaviour.
  • 4 million observations that include both nominal and anomalous segments. This is suitable to evaluate both semi-supervised approaches (novelty detection) as well as unsupervised approaches (outlier detection).
• 200 anomalous segments. One anomalous segment may contain several successive anomalous observations / timestamps. Only the last 4 million observations contain anomalous segments.
• Different types of anomalies to understand what anomaly types can be detected by different approaches.
• Fine control over ground truth. As this is a simulated system with deliberate anomaly injection, the start and end time of the anomalous behaviour is known very precisely. In contrast to real world datasets, there is no risk that the ground truth contains mislabelled segments which is often the case for real data.
• Obvious anomalies. The simulated anomalies have been designed to be "easy" to be detected for human eyes (i.e., there are very large spikes or oscillations), hence also detectable for most algorithms. It makes this synthetic dataset useful for screening tasks (i.e., to eliminate algorithms that are not capable to detect those obvious anomalies). However, during our initial experiments, the dataset turned out to be challenging enough even for state-of-the-art anomaly detection approaches, making it suitable also for regular benchmark studies.
• Context provided. Some variables can only be considered anomalous in relation to other behaviours. A typical example consists of a light and switch pair. The light being either on or off is nominal, the same goes for the switch, but having the switch on and the light off shall be considered anomalous. In the CATS dataset, users can choose (or not) to use the available context, and external stimuli, to test the usefulness of the context for detecting anomalies in this simulation.
• Pure signal ideal for robustness-to-noise analysis. The simulated signals are provided without noise: while this may seem unrealistic at first, it is an advantage since users of the dataset can decide to add on top of the provided series any type of noise and choose an amplitude. This makes it well suited to test how sensitive and robust detection algorithms are against various levels of noise.
• No missing data. You can drop whatever data you want to assess the impact of missing values on your detector with respect to a clean baseline.

[1] Example Benchmark of Anomaly Detection in Time Series: “Sebastian Schmidl, Phillip Wenig, and Thorsten Papenbrock. Anomaly Detection in Time Series: A Comprehensive Evaluation. PVLDB, 15(9): 1779 - 1797, 2022. doi:10.14778/3538598.3538602”

About Solenix

Solenix is an international company providing software engineering, consulting services and software products for the space market. Solenix is a dynamic company that brings innovative technologies and concepts to the aerospace market, keeping up to date with technical advancements and actively promoting spin-in and spin-out technology activities. We combine modern solutions which complement conventional practices. We aspire to achieve maximum customer satisfaction by fostering collaboration, constructivism, and flexibility.

Verbal and Quantitative Reasoning GRE scores and percentiles were collected by querying the student database for the appropriate information. Any student records that were missing data such as GRE scores or grade point average were removed from the study before the data were analyzed. The GRE Scores of entering doctoral students from 2007-2012 were collected and analyzed. A total of 528 student records were reviewed. Ninety-six records were removed from the data because of a lack of GRE scores. Thirty-nine of these records belonged to MD/PhD applicants who were not required to take the GRE to be reviewed for admission. Fifty-seven more records were removed because they did not have an admissions committee score in the database. After 2011, the GRE’s scoring system was changed from a scale of 200-800 points per section to 130-170 points per section. As a result, 12 more records were removed because their scores were representative of the new scoring system and therefore were not able to be compared to the older scores based on raw score. After removal of these 96 records from our analyses, a total of 420 student records remained which included students that were currently enrolled, left the doctoral program without a degree, or left the doctoral program with an MS degree. To maintain consistency in the participants, we removed 100 additional records so that our analyses only considered students that had graduated with a doctoral degree. In addition, thirty-nine admissions scores were identified as outliers by statistical analysis software and removed for a final data set of 286 (see Outliers below). Outliers We used the automated ROUT method included in the PRISM software to test the data for the presence of outliers which could skew our data. The false discovery rate for outlier detection (Q) was set to 1%. After removing the 96 students without a GRE score, 432 students were reviewed for the presence of outliers. ROUT detected 39 outliers that were removed before statistical analysis was performed. Sample See detailed description in the Participants section. Linear regression analysis was used to examine potential trends between GRE scores, GRE percentiles, normalized admissions scores or GPA and outcomes between selected student groups. The D’Agostino & Pearson omnibus and Shapiro-Wilk normality tests were used to test for normality regarding outcomes in the sample. The Pearson correlation coefficient was calculated to determine the relationship between GRE scores, GRE percentiles, admissions scores or GPA (undergraduate and graduate) and time to degree. Candidacy exam results were divided into students who either passed or failed the exam. A Mann-Whitney test was then used to test for statistically significant differences between mean GRE scores, percentiles, and undergraduate GPA and candidacy exam results. Other variables were also observed such as gender, race, ethnicity, and citizenship status within the samples. Predictive Metrics. The input variables used in this study were GPA and scores and percentiles of applicants on both the Quantitative and Verbal Reasoning GRE sections. GRE scores and percentiles were examined to normalize variances that could occur between tests. Performance Metrics. The output variables used in the statistical analyses of each data set were either the amount of time it took for each student to earn their doctoral degree, or the student’s candidacy examination result.