A Comprehensive set of Analytical Engines and Tools
“Based on technology developed at Pattern Discovery Technologies and the University of Waterloo, Discover*e employs pattern recognition algorithms to isolate complex inter-dependencies in the data and uncover meaningful relationships (new knowledge) in complex industrial processes.”
New found knowledge uncovered by Discover*e can be captured in a model to provide powerful predictive insight, enabling more accurate, informed decisions to be made. The technology is very easily embedded into end-user applications, providing easy to use applications that are tailored to any specific environment (for example, an oil sands plant). Because of the “data-centric” nature of this approach, the insights that are uncovered are supported by sound statistical evidence; inherently, the knowledge that is gained is supported by a greater degree of confidence and applied with a greater degree of consistency across the environment.
The suite of applications powered by Discover*e, result in easy to use applications that are interpretative and actionable by process engineers, plant managers, control room operators and maintenance personnel.
Pattern Discovery Technologies is an industry leader in analytics largely because of our commitment to research and development. And so, when it comes to our pattern detection engine, Discover*e, we not only employ industry-standard methods, we’ve developed and patented our own analytical techniques. This ensures that Discover*e offers an unprecedented level of analysis that is unparalleled in the industry.
In order to illustrate this, consider the following table. Each entry represents an analytical method employed by Discover*e. However those marked with an asterisk (*) are exclusive to Pattern Discovery Technologies. That’s because we have researched, developed and patented these new analytical techniques to achieve the most comprehensive pattern-detection engine possible. That engine remains versatile enough to be readily integrated with custom software for your process needs.
Pattern Detection Methods
Those methods marked with an asterisk (*) have been developed by Pattern Discovery Technologies Inc.
Exploratory Data Mining
- High Order Pattern Discovery* Exploratory Data Mining
- Pattern Synthesis and Data Grouping* Exploratory Data Mining
- Subjective Pruning* Exploratory Data Mining
- Differencing Analysis* Exploratory Data Mining
- Pattern Model Browsing* Exploratory Data Mining
- Hyperbolic Data Visualization* Visual Analysis
- Hyperbolic Tree Visualization* Visual Analysis
- Data Matrix Visualization* Visual Analysis
- Cluster Visualization* Visual Analysis
- RouteMap Visualization of Associations* Visual Analysis
- High Order Pattern Discovery based Rule Classifier* Classification
- Decision Tree (C4.5) Classification
- Discrete Valued Clustering* Clustering Analysis
- Density Based Clustering (DBScan) Clustering Analysis
- K Clustering (kMeans, KNN, Fuzzy kMeans, Jarvis-Patrick, etc.) Clustering Analysis
- Hierarchical Clustering Clustering Analysis
- Regression (Principle Component, Partial Least Square, Multiple Linear) Statistical Methods
- Multidimensional Scaling Statistical Methods
- SOM/Kohenen Net Statistical Methods
- Temporal Pattern Discovery from Multiple Time Series* Time Based Analysis
Getting Results Using Discover*e and Association Discovery
Association Discovery is a powerful tool within the Discover*e analysis suite that measures the amount of information contained in a dataset. Specifically, the frequency of occurrence of all possible combinations of factors is compared to the frequency that would be expected to occur by chance. The difference or “residual” tells us that an event is happening far more often or far less often than expected. The statistical significance of these events is ranked in accordance with a corresponding confidence level. The result is a descriptive pattern that can be fully investigated by domain experts to gain a better understanding of these hidden, high-order relationships.
Being able to predict with confidence is a powerful asset in any decision making process. To facilitate confident predictions, the descriptive patterns found are transformed into production rules. The strength of a rule is measured by the Weight of Evidence (WoE) supporting it, which provides statistical proof that a pattern contributes to a specific output or result. Following this same approach, new inputs into the process can be classified using the established rules.
This approach provides a completely unbiased, data driven approach to data mining and predictive analytics. It works with multi-dimensional relationships, taking into account all of the factors that should be considered in performing analyses of complex datasets. The results of both the data mining exercise and the predictive models are descriptive in nature, meaning they can be interpreted in plain English.
Knowing the reasons why a particular prediction is being made can establish a proper set of actions or decisions based on the anticipated result. Unlike many other types of learning systems, Discover*e exposes the underlying behaviours in an accessible manner that allows plant engineers to make informed decisions and build effective process control and maintenance strategies, anticipate disruptive events before they occur, and drive continuous improvement.