Archive for the ‘Knowledge Systems’ category

Thoughts on Data Mining

March 10, 2012

Data mining (sometimes called data or knowledge discovery) is the process of analyzing data from different perspectives and summarizing it into useful information (see prior blogs including The Data Information Hierarchy series). The term is overused and conjures impressions that do not reflect the true state of the industry. Knowledge Discovery from Databases (KDD) is more descriptive and not as misused – but the base meaning is the same.

Nevertheless, this definition of data mining is a very general definition and does not convey the different aspects of data mining / knowledge discovery.

The basic types of Data Mining are:

  • Descriptive data mining, and
  • Predictive data mining

Descriptive Data Mining generally seeks groups, subgroups and clusters. Algorithms are developed that draw associative relationships from which actionable results may be derived. (ie. a diamond head snake should be considered poisonous.)

Generally, a descriptive data mining result will appear as a series of if – then – elseif – then … conditions. Alternatively, a system of scoring may be used much like some magazine based self assessment exams. Regardless of the approach, the end result is a clustering of the samples with some measure of quality.

Predictive Data Mining is then performing an analysis on previous data to derive a prediction to the next outcome. For example: new business incorporation tend to look for credit card merchant solutions. This may seem obvious, but someone had to discover this tendency – and then exploit it.

Data mining is ready for application in the business community because it is supported by three technologies that are now sufficiently mature: 1) massive data collection, 2) powerful multiprocessor computers, and 3) data mining algorithms (http://www.thearling.com/text/dmwhite/dmwhite.htm).

Kurt Thearling identifies five type od data mining: (definitions taken from Wikipedia)

A decision tree is a decision support tool that uses a tree-like graph or model of decisions and their possible consequences, including chance event outcomes, resource costs, and utility. Decision trees are commonly used in operations research, specifically in decision analysis, to help identify a strategy most likely to reach a goal. If in practice decisions have to be taken online with no recall under incomplete knowledge, a decision tree should be paralleled by a Probability model as a best choice model or online selection model algorithm. Another use of decision trees is as a descriptive means for calculating conditional probabilities.

Nearest neighbour or shortest distance is a method of calculating distances between clusters in hierarchical clustering. In single linkage, the distance between two clusters is computed as the distance between the two closest elements in the two clusters.

The term neural network was traditionally used to refer to a network or circuit of biological neurons. The modern usage of the term often refers to artificial neural networks, which are composed of artificial neurons or nodes.

Rule induction is an area of machine learning in which formal rules are extracted from a set of observations. The rules extracted may represent a full scientific model of the data, or merely represent local patterns in the data.

Cluster analysis or clustering is the task of assigning a set of objects into groups (called clusters) so that the objects in the same cluster are more similar (in some sense or another) to each other than to those in other clusters.

Advertisements

Artificial Intelligence vs Algorithms

February 9, 2012

I first considered aspects of artificial intelligence (AI) in the 1980s while working for General Dynamics as an Avionics Systems Engineer on the F-16. Over the following 3 decades, I continued to follow the concept until I made a realization – AI is just an algorithm. Certainly the goals of AI will one day be reached, but the manifestation metric of AI is not well defined.

Mark Reynolds is currently at Southwestern Energy where he works in the Fayetteville Shale Drilling group as a Staff Drilling Data Analyst. In this position, he pulls his experiences in data processing, data analysis, and data presentation to improve Southwestern Energy’s work in the natural gas production and mid-stream market.

Recently, Mark has been working toward improved data collection, retention, and utilization in the real-time drilling environment.

www.ProfReynolds.com

Consider the Denver International Airport. The baggage handling system was state of the art, touted as AI based and caused the delay of the opening by 16 months and cost $560M to fix. (more – click here) In the end, the entire system was replaced with a more stable system based not on a learning or deductive system, but upon much more basic routing and planning algorithms which may be deterministically designed and tested.

Consider the Houston traffic light system. Mayors have been elected on the promise to apply state of the art computer intelligence. Interconnected traffic lights, traffic prediction, automatic traffic redirection. Yet the AI desired results in identifiable computer algorithms with definitive behavior and expectations. Certainly an improvement, but not a thinking machine. The closest thing to automation is the remote triggering features used by the commuter rail and emergency vehicles.

So algorithms form the basis for computer advancement. And these algorithms may be applied with human interaction to learn the new lessons so necessary to achieving behavioral improvement with the computers. Toward this objective, distinct fields of study are untangling interrelated elements – clustering, neural networks, case based reasoning, and predictive analytics are just a few.

When AI can be achieved, it will be revolutionary. But until that time, deterministic algorithms, data mining, and predictive analytics will be at the core of qualitative and quantitative advancement.

Predictive Analytics

September 9, 2011

Predictive analytics is used in actuarial science, financial services, insurance, telecommunications, retail, travel, healthcare, pharmaceuticals and other fields (Wikipedia). But operations – manufacturing, processing, etc., have been a little slower to encompass the concept. A drilling engineer friend of mine says “just put my hand on the brake lever and I’ll drill that well”. He probably can, but few of the rest of us can, or want to.

We want to see operating parameters, performance metrics, and process trends. All this because we want to have the necessary information and knowledge to assimilate understanding and invoke our skill set (wisdom). In this scenario, we are responding to stimulus, we are applying “reactive analytics”. But systems get more complex, operations becomes more intertwined, performance expectations become razor-thin. And with this complexity grows demand for better assistance from technology. In this case, the software performs the integrated analysis and the results is “predictive analytics”. And with predictive analytics comes the close cousin: decision models and decision trees.

Sullivan McIntyre, in his article From Reactive to Predictive Analytics, makes an observation about predictive analytics in social media that is mirrored in operations:

There are three key criteria for making social data useful for making predictive inferences:

  • Is it real-time? (Or as close to real-time as possible)
  • Is it metadata rich?
  • Is it integrated?

Having established these criteria, the nature of the real-time data and the migration of historical data into real-time, predictive analytics becomes achievable.

The Value of Real-Time Data

August 31, 2011

Real-Time data is a challenge to any process-oriented operation. But the functionality of the data is difficult to describe in such a way that team members not well versed in data management. Toward that end, four distinct phases of data have been identified:

  1. Real-Time: streaming data
    visualized and considered – system responds
  2. Forensic: captured data
    archived, condensed – system learns
  3. Data Mining: consolidated data
    hashed and clustered – system understands
  4. Predictive Analytics: patterned data
    compared and matched – system anticipates

A mnore detailed explanation of these phases may be:

Control and Supervision: Real-time data is used to provide direct HMI (human-machine-interface) and permit the human computer to monitor / control the operations from his console. The control and supervision phase of real-time data does not, as part of its function, record the data. (However, certain data logs may be created for legal or application development purposes.) Machine control and control feedback loops require, as a minimum, real-time data of sufficient quality to provide steady operational control.

Forensic Analysis and Lessons Learned: Captured data (and, to a lesser extent, data and event logs) are utilized to investigate specific performance metrics and operations issues. Generally, this data is kept in some form for posterity, but it may be filtered, processed, or purged. Nevertheless, the forensic utilization does represent post-operational analytics. Forensic analysis is also critical to prepare an operator for an upcoming similar process – similar in function, geography, or sequence.

Data Mining: Data mining is used to research previous operational events to locate trends, areas for improvement, and prepare for upcoming operations. Data mining is used identify a bottleneck or problem area as well as correlate events that are less than obvious.

Proactive / Predictive Analytics: The utilization of data streams, both present and previous, in an effort to predict the immediate (or distant) future requires historical data, data mining, and the application of learned correlations. Data mining may provide correlated events and properties, but the predictive analytics will provide the conversion of the correlations into positive, immediate performance and operational changes. (This utilization is not, explicitly AI, artificial intelligence, but the two are closely related)

The Data-Information Hierarcy, Part 3

August 31, 2011

The Data-Information Hierarchy is frequently represented as
Data –> Information –> Knowledge –> Understanding –> Wisdom.

Or it is sometimes shortened to 4 steps, omitting Understanding. But, in fact, there are two predecessor steps: chaos and symbol. These concepts have been discussed in prior blogs (https://profreynolds.wordpress.com/2011/01/31/the-data-information-hierarcy/
https://profreynolds.wordpress.com/2011/02/11/the-data-information-hierarcy-part-2/).

Chaos is that state of lack of understanding best compared to the baby first perceiving the world around him. There is no comprehension of quantities or values, but a perception of large and small.

Symbol (or symbolic representation) represents the first stages of quantification. As such, symbolic representation and quantification concepts from the predecessor to Data.

So the expanded Data-Information hierarchy is represented in the seven steps:

Chaos –>
          Symbol –>
                    Data –>
                              Information –>
                                        Knowledge –>
                                                  Understanding –>
                                                            Wisdom

Continuing with this Data-Hierarchy paradigm, we can represent the five primary steps with the simple explanation:

  • Data and Information : ‘Know What’
  • Knowledge : ‘Know How’
  • Understanding : ‘Know Why’
  • Wisdom : ‘Use It’

The Big Crew Change

May 17, 2011

“The Big Crew Change” is an approaching event within the oil and gas industry when the mantle of leadership will move from the “calculators and memos” generation to the “connected and Skype” generation. In a blog 4 years ago, Rembrandt observes:

“The retirement of the workforce in the industry is normally referred to as “the big crew change”. People in this sector normally retire at the age of 55. Since the average age of an employee working at a major oil company or service company is 46 to 49 years old, there will be a huge change in personnel in the coming ten years, hence the “big crew change”. This age distribution is a result of the oil crises in ‘70s and ‘80s as shown in chart 1 & 2 below. The rising oil price led to a significant increase in the inflow of petroleum geology students which waned as prices decreased.”

Furthermore, a Society of Petroleum Engineers study found:

“There are insufficient personnel or ‘mid-carrers’ between 30 and 45 with the experience to make autonomous decisions on critical projects across the key areas of our business: exploration, development and production. This fact slows the potential for a safe increase in production considerably”

A study undertaken by Texas Tech University make several points about the state of education and the employability of graduates during this crew change:

  • Employment levels at historic lows
  • 50% of current workers will retire in 6 years
  • Job prospects: ~100% placement for the past 12 years
  • Salaries: Highest major in engineering for new hires

The big challenge: Knowledge Harvesting. “The loss of experienced personnel combined with the influx of young employees is creating unprecedented knowledge retention and transfer problems that threaten companies’ capabilities for operational excellence, growth, and innovation.” (Case Study: Knowledge Harvesting During the Big Crew Change).

In a blog by Otto Plowman, “Retaining knowledge through the Big Crew Change”, we see that

“Finding a way to capture the knowledge of experienced employees is critical, to prevent “terminal leakage” of insight into decisions about operational processes, best practices, and so on. Using of optimization technology is one way that producers can capture and apply this knowledge.When the retiring workforce fail to convey the important (critical) lessons learned, the gap is filled by data warehouses, knowledge systems, adaptive intelligence, and innovation.”

When the retiring workforce fail to convey the important (critical) lessons learned, the gap is filled by data warehouses, knowledge systems, adaptive intelligence, and innovation. Perhaps the biggest challenge is innovation. Innovation will drive the industry through the next several years. Proactive intelligence, coupled with terabyte upon terabyte of data will form the basis.

The future: the nerds will take over from the wildcatter.


%d bloggers like this: