Tag: Machine Learning

Adv DBs: Unsupervised and Supervised Learning

Unsupervised and Supervised Learning:

Supervised learning is a type of machine learning that takes a given set of data points, we need to choose a function that gives users a classification or a value.  So, eventually, you will get data points that no longer defines a classification or a value, thus the machine now has to solve for that function. There are two main types of supervised learning: Classification (has a finite set, i.e. based on person’s chromosomes in a database, their biological gender is either male or female) and Regression (represents real numbers in the real space or n-dimensional real space).  In regression, you can have a 2-dimensional real space, with training data that gives you a regression formula with a Pearson’s correlation number r, given a new data point, can the machine use the regression formula with correlation r to predict where that data point will fall on in the 2-dimensional real space (Mathematicalmonk’s channel, 2011a).

Unsupervised learning aims to uncover homogenous subpopulations in databases (Connolly & Begg, 2015). In Unsupervised learning you are given data points (values, documents, strings, etc.) in n-dimensional real space, the machine will look for patterns through either clustering, density estimation, dimensional reduction, etc.  For clustering, one could take the data points and placing them in bins with common properties, sometimes unknown to the end-user due to the vast size of the data within the database.  With density estimation, the machine is fed a set of probability density functions to fit the data and it begins to estimates the density of that data set.  Finally, for dimensional reduction, the machine will find some lower dimensional space in which the data can be represented (Mathematicalmonk’s channel, 2011b).  With the dimensional reduction, it can destroy the structure that can be seen in the higher-order dimensions.

Applications suited to each method

  • Supervised: defining data transformations (Kelvin to Celsius, meters per second to miles per hour, classifying a biological male or female given the number of chromosomes, etc.), predicting weather (given the initial & boundary conditions, plug them into formulas that predict what will happen in the next time step).
  • Unsupervised: forecasting stock markets (through patterns identified in text mining news articles, or sentiment analysis), reducing demographical database data to common features that can easily describe why a certain population will fit a result over another (dimensional reduction), cloud classification dynamical weather models (weather models that use stochastic approximations, Monte Carlo simulations, or probability densities to generate cloud properties per grid point), finally real-time automated conversation translators (either spoken or closed captions).

Most important issues related to each method

Unsupervised machine learning is at the bedrock of big data analysis.  We could use training data (a set of predefined data that is representative of the real data in all its n-dimensions) to fine-tune the most unsupervised machine learning efforts to reduce error rates (Barak & Modarres, 2015). What I like most about unsupervised machine learning is its clustering and dimensional reduction capabilities, because it can quickly show me what is important about my big data set, without huge amounts of coding and testing on my end.

References:

Adv Topics: Extracting Knowledge from big data

The evolution of data to wisdom is defined by the DIKW pyramid, where Data is just facts without any context, but when facts are used to understand relationships it generates Information (Almeyer-Stubbe & Coleman, 2014). That information can be used to understand patterns, it can then help build Knowledge, and when that knowledge is used to understand principles, it builds Wisdom (Almeyer-Stubbe & Coleman, 2014; Bellinger, Castro, Mills, n.d.). Building an understanding to jump from one level of the DIKW pyramid, is an appreciation of learning “why” (Bellinger et al., n.d.). Big data was first coined in a Gartner blog post, is data that has high volume, variety, and velocity, but without any interest in understanding that data, data scientist will lack context (Almeyer-Stubbe & Coleman, 2014; Bellinger et al., n.d.; Laney, 2001). Therefore, applying the DIKW pyramid can help turn that big data into extensive knowledge (Almeyer-Stubbe & Coleman, 2014; Bellinger et al., n.d.; Sakr, 2014). Extensive knowledge is a derived from placing meaning to big data usually in the form of predictive analytics algorithms (Sakr, 2014).

Machine learning requires historical data and is part of the data analytics process under data mining to understand hidden patterns or structures within the data (Almeyer-Stubbe & Coleman, 2014). Machine learning is easier to build and maintain than other classical data mining techniques (Wollan, Smith, & Zhou, 2010). Machine learning algorithms include clustering, classification, and association rules techniques and the right algorithm from any of these three techniques must be selected that meet the needs of the data (Services, 2015). Unsupervised machine learning techniques like clustering are used when data scientist do not understand or classify data prior to data mining techniques to understand hidden structures within the data set (Brownlee, 2016; Services, 2015). Supervised machine learning involves model training and model testing to aid in understanding which input variables feed into an output variable, involving such techniques as classification and regression (Brownlee, 2016).

An example of an open source Hadoop machine learning algorithm library would include Apache Mahout, which can be found at http://mahout.apache.org (Lublinsky, Smith, & Yakubovich, 2013). A limitation from learning from historical data to predict the future is it can “stifle innovation and imagination” (Almeyer-Stubbe & Coleman, 2014). Another limitation can exist that current algorithms may not run on distributed database systems. Thus some tailoring of the algorithms may be needed (Services, 2015). The future of machine learning involves its algorithms becoming more interactive to the end user, known as active learning (Wollan, Smith, & Zhou, 2010).

Case Study: Machine learning, medical diagnosis, and biomedical engineering research – commentary (Foster, Koprowski, & Skufca, 2014)

The authors created a synthetic training data set to simulate a typical medical classification problem of healthy and ill people and assigned random numbers to 10 health variables. Given this information, the actual classification accuracy should be 50%, which is also similar to pure chance alone. These authors found that when classification machine learning algorithms are misapplied, it can lead to false results. This was proven when their model took only 50 people to produce similar accuracy values of pure chance alone. Thus, the authors of this paper were trying to warn the medical field that misapplying classification techniques can lead to overfitting.

The authors then looked at feature selection for classifying Hashimoto’s disease from 250 clinical ultrasound data with the disease and 250 healthy people. Ten variables were selected to help classify these images, and a MATLAB machine learning algorithm was trained on 400 people (200 healthy and 200 ill) to then be tested on 100 people (50 healthy and 50 ill). They were able to show that when 3-4 variables were used they produced better classification results, thus 3-4 variables had huge information gain. This can mislead practitioners, because of the small data set that could be generalized too broadly and the lack of independence between training and testing datasets. The authors argued that larger data sets are needed to get rid of some of the issues that could result in the misapplication of classifiers.

The authors have the following four recommendations when considering the use of supervised machine learning classification algorithms:

    1. Clearly, state the purpose of the study and come from a place of understanding of that problem and its applications.
    2. Minimize the number of a variable when used in classifiers, such as using pruning algorithms in classification algorithms to only select certain variables that meet a certain level of information gain. This is more important with smaller data sets than with big data.
    3. Understand that classifiers are sensitive and that results gained from one set of instances might require further adjustments to be implemented elsewhere.
    4. Classification algorithms and data mining are part of the experimental process not the answer to all problems.

Resources:

Fraud detection in the health care industry using analytics

Fraud is deception, fraud detection is really needed, because as fraud detection algorithms are improving, the rate of fraud is increasing (Minelli, Chambers, &, Dhiraj, 2013). Hadoop and the HFlame distribution have to be used to help identify fraudulent data in other companies like banking in near-real-time (Lublinsky, Smith, & Yakubovich, 2013).

Data mining has allowed for fraud detection via multi-attribute monitoring, where it tries to find hidden anomalies by identifying hidden patterns through the use of class description and class discrimination (Brookshear & Brylow, 2014; Minellli et al., 2013). Class Descriptions identify patterns that define a group of data, and class discrimination identifies patterns that divide groups of data (Brookshear & Brylow, 2014). As data flows in, data is monitored through validity check and detection rules and gives them a score, such that if the validity and detection score surpasses a threshold, that data point is flagged as potentially suspicious (Minelli et al., 2013).

This is a form of outlier data mining analysis, where data that doesn’t fit any of the above groups of data that has been described and discriminated can be used to identify fraudulent data (Brookshear & Brylow, 2014; Connolly & Begg, 2014). Minelli et al. (2013), stated that using historical data to build up the validity check and detection rules with real-time data can help identify outliers in near-real time. However, what about predicting fraud?  In the future, companies will be using Hadoop’s machine learning capability paired with its fraud detection algorithms to provided predictive modeling of fraud events (Lublinsky, Smith, & Yakubovich, 2013).

A process mining framework for the detection of healthcare fraud and abuse case study (Yang & Hwang, 2006)

Fraud exists in processing health insurance claims because there are more opportunities to commit fraud because there are more channels of communication: service providers, insurance agencies, and patients. Any one of these three people can commit fraud, and the highest chance of fraud occurs where service providers can do unnecessary procedures putting patients at risk. Thus this case study provided the framework on how to conduct automated fraud detection. The study collected data from 2543 gynecology patients from 2001-2002 from a hospital, where they filtered out noisy data, identified activities based on medical expertise, identified fraud in about 906.

Before data mining and machine learning, the process was heavily reliant on medical professional with subject matter expertise to detect fraud, which was costly for multiple resources.  Also, machine learning is not subject to human and manual error that is common with humans.  Machine learning algorithms for fraud detection relies on clinical pathways, which are defined as the right people giving the right care services in the right order, with the aim at the reduction of waste and implementing best practices.  Any deviation from this that is abnormal can be flagged by the machine learning algorithm.

References

  • Brookshear, G., & Brylow, D. (2014). Computer Science: An Overview, (12th). Pearson Learning Solutions. VitalBook file.
  • Connolly, T., Begg, C. (2014). Database Systems: A Practical Approach to Design, Implementation, and Management, (6th). Pearson Learning Solutions. VitalBook file.
  • Lublinsky, B., Smith, K., & Yakubovich, A. (2013). Professional Hadoop Solutions. Wrox. VitalBook file.
  • Minelli, M., Chambers, M., &, Dhiraj, A. (2013). Big Data, Big Analytics: Emerging Business Intelligence and Analytic Trends for Today’s Businesses. John Wiley & Sons P&T. VitalBook file.
  • Yang, W. S., & Hwang, S. Y. (2006). A process-mining framework for the detection of healthcare fraud and abuse.Expert Systems with Applications, 31(1), 56-68.

Diagnosis of illness via big data

Big data is defined as high volume, high variety/complexity, and high velocity, which is known as the 3Vs (Services, 2015). Using Machine Learning and Artificial Intelligence, they do well at analyzing patterns from frequent and voluminous amounts of data at faster speeds than humans, but they fail to recognize patterns in infrequent and small amounts of data (Goldbloom, 2016). Thus, the use of data analytic theories and techniques on big data rather than novel situations in healthcare is vital to understand.

Fayyad, Piatetsky-Shapiro, and Smyth (1996) defined that data analytics can be divided into descriptive and predictive analytics. Vardarlier and Silaharoglu (2016) agreed with Fayyad et al. (1996) division but added prescriptive analytics.  Thus, these three divisions of big data analytics are:

  • Descriptive analytics explains “What happened?”
  • Predictive analytics explains “What will happen?”
  • Prescriptive analytics explains “Why will it happen?”

(Fayyad et al., 1996; Vardarlier & Silahtaroglu, 2016). Depending on the goal of diagnosing illnesses with the use of big data analytics should depend on the theory/division one should choose.  Raghupathi & Raghupathi (2014), stated some common examples of big data in the healthcare field to be: personal medical records, radiology images, clinical trial data, 3D imaging, human genomic data, population genomic data, biometric sensor reading, x-ray films, scripts, and traditional paper files.

The use of big data analytics to understand the 23 pairs of chromosomes that are the building blocks for people. Healthcare professionals are using the big data generated from our genomic code to help predict which illnesses a person could get (Services, 2013). Thus, using predictive analytics tools and algorithms like decision trees would be of some use.  Another use of predictive analytics and machine learning can be applied to diagnosing an eye disease like diabetic retinopathy from an image by using classification algorithms (Goldbloom, 2016).

The study of epigenetics, which are what parts of the genetic code is turned on versus turned off, can help explain why will certain illnesses are more probable to occur in the future (What is epigenetics, n.d.).  Thus, the use of prescriptive analytics could be of some use in the study of epigenetics. Currently, clinical trials use descriptive analytics to help calculate true positives, false positives, true negatives, and false negatives of a drug treatment versus a placebo are commonly used.  Thus, depending on the goal of diagnosing illnesses and the problem, that should help define which theories and techniques of big data analytics to use. The use of different data analytics techniques and theories based on the problem and data can change how healthcare jobs in the next 30 years from today (Goldbloom, 2016; McAfee, 2013).

References

Compelling topics

Hadoop, XML and Spark

Hadoop is predominately known for its Hadoop Distributed File System (HDFS) where the data is distributed across multiple systems and its code for running MapReduce tasks (Rathbone, 2013). MapReduce has two queries, one that maps the input data into a final format and split across a group of computer nodes, while the second query reduces the data in each node so that when combining all the nodes it can provide the answer sought (Eini, 2010).

XML documents represent a whole data file, which contains markups, elements, and nodes (Lublinsky, Smith, & Yakubovich,, 2013; Myer, 2005):

  • XML markups are tags that helps describe the data start and end points as well as the data properties/attributes, which are encapsulated by < and a >
  • XML elements are data values, encapsulated by an opening <tag> and a closing </tag>
  • XML nodes are part of the hierarchical structure of a document that contains a data element and its tags

Unfortunately, the syntax and tags are redundant, which can consume huge amounts of bytes, and slow down processing speeds (Hiroshi, 2007)

Five questions must be asked before designing an XML data document (Font, 2010):

  1. Will this document be part of a solution?
  2. Will this document have design standards that must be followed?
  3. What part may change over time?
  4. To what extent is human readability or machine readability important?
  5. Will there be a massive amount of data? Does file size matter?

All XML data documents should be versioned, and key stakeholders should be involved in the XML data design process (Font, 2010).  XML is a machine and human readable data format (Smith, 2012). With a goal of using XML for MapReduce, we need to assume that we need to map and reduce huge files (Eini, 2010; Smith 2012). Unfortunately, XML doesn’t include sync markers in the data format and therefore MapReduce doesn’t support XML (Smith, 2012). However, Smith (2012) and Rohit (2013) used the XmlInputFormat class from mahout to work with XML input data into HBase. Smith (2012), stated that the Mahout’s code needs to know the exact sequence of XML start and end tags that will be searched for and Elements with attributes are hard for Mahout’s XML library to detect and parse.

Apache Spark started from a working group inside and outside of UC Berkley, in search of an open-sourced, multi-pass algorithm batch processing model of MapReduce (Zaharia et al., 2012). Spark is faster than Hadoop in iterative operations by 25x-40x for really small datasets, 3x-5x for relatively large datasets, but Spark is more memory intensive, and speed advantage disappears when available memory goes down to zero with really large datasets (Gu & Li, 2013).  Apache Spark, on their website, boasts that they can run programs 100X faster than Hadoop’s MapReduce in Memory (Spark, n.d.). Spark outperforms Hadoop by 10x on iterative machine learning jobs (Gu & Li, 2013). Also, Spark runs 10x faster than Hadoop on disk memory (Spark, n.d.). Gu and Li (2013), recommend that if speed to the solution is not an issue, but memory is, then Spark shouldn’t be prioritized over Hadoop; however, if speed to the solution is critical and the job is iterative Spark should be prioritized.

Data visualization

Big data can be defined as any set of data that has high velocity, volume, and variety, also known as the 3Vs (Davenport & Dyche, 2013; Fox & Do, 2013; Podesta, Pritzker, Moniz, Holdren, & Zients, 2014).  What is considered to be big data can change with respect to time.  What is considered as big data in 2002 is not considered big data in 2016 due to advancements made in technology over time (Fox & Do, 2013).  Then there is Data-in-motion, which can be defined as a part of data velocity, which deals with the speed of data coming in from multiple sources as well as the speed of data traveling between systems (Katal, Wazid, & Goudar, 2013). Essentially data-in-motion can encompass data streaming, data transfer, or real-time data. However, there are challenges and issues that have to be addressed to conducting real-time analysis on data streams (Katal et al., 2013; Tsinoremas et al., n.d.).

It is not enough to analyze the relevant data for data-driven decisions but also selecting relevant visualizations of that data to enable those data-driven decision (eInfochips, n.d.). There are many types of ways to visualize the data to highlight key facts through style and succinctly: tables and rankings, bar charts, line graphs, pie charts, stacked bar charts, tree maps, choropleth maps, cartograms, pinpoint maps, or proportional symbol maps (CHCF, 2014).  The above visualization plots, charts, maps and graphs could be part of an animated, static, and Interactive Visualizations and would it be a standalone image, dashboards, scorecards, or infographics (CHCF, 2014; eInfochips, n.d.).

Artificial Intelligence (AI)

Artificial Intelligence (AI) is an embedded technology, based off of the current infrastructure (i.e. supercomputers), big data, and machine learning algorithms (Cyranoski, 2015; Power, 2015). AI can provide tremendous value since it builds thousands of models and correlations automatically in one week, which use to take a few quantitative data scientist years to do (Dewey, 2013; Power, 2015).  Unfortunately, the rules created by AI out of 50K variables lack substantive human meaning, or the “Why” behind it, thus making it hard to interpret the results (Power, 2015).

“Machines can excel at frequent high-volume tasks. Humans can tackle novel situations.” said by Anthony Goldbloom. Thus, the fundamental question that decision makers need to ask, is how the decision is reduced to frequent high volume task and how much of it is reduced to novel situations (Goldbloom, 2016).  Therefore, if the ratio is skewed on the high volume tasks then AI could be a candidate to replace decision makers, if the ratio is evenly split then AI could augment and assist decision makers, and if the ratio is skewed on novel situations, then AI wouldn’t help decision makers.  They novel situations is equivalent to our tough challenges today (McAfee, 2013).  Finally, Meetoo (2016), warned that it doesn’t matter how intelligent or strategic a job could be, if there is enough data on that job to create accurate rules it can be automated as well; because machine learning can run millions of simulations against itself to generate huge volumes of data to learn from.

 

Resources:

Data Tools: Artificial Intelligence and Data Analytics

Machine learning, also known as Artificial Intelligence (AI) adds an intelligence layer to big data to handle the bigger sets of data to derive patterns from it that even a team of data scientist would find challenging (Maycotte, 2014; Power, 2015). AI makes their insights not by how machines are programmed, but how the machines perceive the data and take actions from that perception, essentially conducting self-learning (Maycotte, 2014).  Understanding how a machine perceives the big dataset is a hard task, which also makes it hard to interpret the resulting final models (Power, 2015).  AI is even revolutionizing how we understand what intelligence is (Spaulding, 2013).

So what is intelligence

At first, doing arithmetic was thought of as a sign of biological intelligence until the invention of the digital computers, which then shift biological intelligence to be known for logical reasoning, deduction and inferences to eventually fuzzy logic, grounded learning, and reasoning under uncertainty, which is now matched through Bayes Nets probability and current data analytics (Spaulding, 2013). So as humans keep moving the dial of what biological intelligence is to a more complex structure, if it requires high frequency and voluminous data, then it can be matched by AI (Goldbloom, 2016).  Therefore, as our definition of intelligence expands so will drive the need to capture intelligence artificially, driving change in how big datasets are analyzed.

AI on influencing the future of data analytics modeling, results, and interpretation

This concept should help revolutionize how data scientists and statisticians think about which hypotheses to ask, which variables are relevant, how do the resulting outputs fit in an appropriate conceptual model, and why do these patterns hidden in the data help generate the decision outcome forecasted by AI (Power, 2015). To figure out or make sense of these models would require subject matter experts from multiple fields and multiple levels of employment hierarchy analyzing these model outputs because it is through diversity and inclusion of thought will we understand an AI’s analytical insight.

Also, owning data is different from understanding data (Lapowsky, 2014). Thus, AI can make use of data hidden in “dark wells” and silos, where the end-user had no idea that the data even existed, to begin with, which allows for a data scientist to gain a better understanding of their datasets (Lapowsky, 2014; Power, 2015).

AI on generating datasets and using data analytics for self-improvements

Data scientists currently collected, preprocess, process and analyze big volumes of data regularly to help provide decision makers with insights from the data to make data-driven decisions (Fayyad, Piatetsky-Shapiro, & Smyth, 1996).  From these data-driven decisions, data scientist then measure the outcomes to prove the effectiveness of their insights (Maycotte, 2014).   This analysis on how the results of data-driven decisions, will allow machine learning algorithms to learn from their decisions and actions to create better ways of searching for key patterns in bigger and future datasets. This is an ability of AI to conduct self-learning based off of the results of data analytics through the use of data analytics (Maycotte, 2014). Meetoo (2016), stated that if there is enough data to create accurate rules it is enough to create insights; because machine learning can run millions of simulations against itself to generate huge volumes of data to which to learn from.

AI on Data Analytics Process

AI is a result of the massive amounts of data being collected, the culmination of ideas from the most brilliant computer scientists of our time, and on an IT infrastructure that didn’t use to exist a few years ago (Power, 2015).  Given that data analytics processes include collecting data, preprocessing data, processing data, and analyzing the results, any improvements made for AI on the infrastructure can have an influence on any part of the data analytics process (Fayyad et al., 1996; Power, 2015).  For example, as AI technology begins to learn how to read raw data to turn that into information, the need for most of the current preprocessing techniques for data cleaning could disappear (Minelli, Chambers, & Dhiraj, 2013). Therefore, as AI begins to advance, newer IT infrastructures will be dreamt up and built such that data analytics and its processes can now leverage this new infrastructure, which can also change the way on how big datasets are analyzed.

Resources:

Data Tools: Artificial Intelligence and Decision Making

“Machines can excel at frequent high-volume tasks. Humans can tackle novel situations.” – Anthony Goldbloom

Jobs today will look drastically different in 30 years from now (Goldbloom, 2016; McAfee, 2013).  Artificial intelligence (AI) works on Sundays, they don’t take holidays, and they work well at high frequency and voluminous tasks, and thus they have the possibility of replacing many of the current jobs of 2016 (Goldbloom, 2016; Meetoo, 2016).  AI has been doing things that haven’t been done before: understanding, speaking, hearing, seeing, answering, writing, and analyzing (McAfee, 2013). Also, AI can make use of data hidden in “dark wells” and silos, where the end-user had no idea that the data even existed, to begin with (Power, 2015). Eventually, AI and machine learning will be commonly used as a tool to augment or replace decision makers.  Goldbloom (2016) gave the example that a teacher may be able to read 10,000 essays or an ophthalmologist may see 50,000 eyes over a 40-year period; whereas a machine can read millions of essays and see millions of eyes in minutes.

Machine learning is one of the most powerful branches to AI, where machines learn from data, similar to how humans learn to create predictions of the future (Cringely, 2013; Cyranoski, 2015; Goldbloom, 2016; Power, 2015). It would take many scientists to analyze a big dataset in its entirety without a loss of memory such that to gain insights and to fully understand how the connections were made in the AI system (Cringely, 2013; Goldbloom, 2016). This is no easy task because the eerily accurate rules created by AI out of thousands of variables can lack substantive human meaning, making it hard to interpret the results and make an informed data-driven decision (Power, 2015).

AI has been used to solve problems in industry and academia already, which has given data scientist knowledge on the current limitations of AI and whether or not they can augment or replace key decision makers (Cyranoski, 2015; Goldbloom, 2016). Machine learning and AI does well at analyzing patterns from frequent and voluminous amounts of data at faster speeds than humans, but they fail to recognize patterns in infrequent and small amounts of data (Goldbloom, 2016).  Therefore, for small datasets artificial intelligence will not be able to replace decision makers, but for big datasets, they would.

Thus, the fundamental question that decision makers need to ask is how is the decision reduced to frequent high volume task and how much of it is reduced to novel situations (Goldbloom, 2016).  Thus, if the ratio is skewed on the high volume tasks then AI could be a candidate to replace decision makers, if the ratio is evenly split, then AI could augment and assist decision makers, and if the ratio is skewed on novel situations, then AI wouldn’t help decision makers.  They novel situations are equivalent to our tough challenges today (McAfee, 2013).

Finally, Meetoo (2016), warned that it doesn’t matter how intelligent or strategic a job could be, if there is enough data on that job to create accurate rules it can be automated as well; because machine learning can run millions of simulations against itself to generate huge volumes of data to learn from.  This is no different than humans doing self-study and continuous practice to be subject matter experts in their field. But people in STEAM (Science, Technology, Engineering, Arts, and Math) will be best equip them for the future world with AI, because it is from knowing how to combine these fields that novel, infrequent, and unique challenges will arise that humans can solve and machine learning cannot (Goldbloom, 2016; McAfee, 2013; Meetoo, 2016).

Resources:

Data Tools: Artificial Intelligence

Big data Analytics and Artificial Intelligence

Artificial Intelligence (AI) is an embedded technology, based off of the current infrastructure (i.e. supercomputers), big data, and machine learning algorithms (Cyranoski, 2015; Power, 2015). Though previously, AI wasn’t able to come into existence without the proper computational power that is provided today (Cringely, 2013).  AI can make use of data hidden in “dark wells” and silos, where the end-user had no idea that the data even existed, to begin with (Power, 2015).  The goal of AI is to use huge amounts of data to draw out a set of rules through machine learning that will effectively replace experts in a certain field (Cringely, 2013; Power, 2015). Cringely (2013) stated that in some situations big data can eliminate the need for theory and that AI can aid in analyzing big data where theory is either lacking or impossible to define.

AI can provide tremendous value since it builds thousands of models and correlations automatically in one week, which use to take a few quantitative data scientist years to do (Dewey, 2013; Power, 2015).  The thing that has slowed down the progression of AI in the past was the creation of human readable computer languages like XML or SQL, which is not intuitive for computers to read (Cringely, 2013).  Fortunately, AI can easily use structured data and now use unstructured data thanks to everyone who tags all these unstructured data either in comments or on the data point itself, speeding up the computational time (Cringely, 2013; Power, 2015).  Dewey (2013), hypothesized that not only will AI be able to analyze big data at speeds faster than any human can, but that the AI system can also begin to improve its search algorithms in phenomena called intelligence explosion.  Intelligence explosion is when an AI system begins to analyze itself to improve itself in an iterative process to a point where there is an exponential growth in improvement (Dewey, 2013).

Unfortunately, the rules created by AI out of 50K variables lack substantive human meaning, or the “Why” behind it, thus making it hard to interpret the results (Power, 2015).  It would take many scientists to analyze the same big data and analyze it all, to fully understand how the connections were made in the AI system, which is no longer feasible (Cringely, 2013).  It is as if data scientist is trying to read the mind of the AI system, and they currently cannot read a human’s mind. However, the results of AI are becoming accurate, with AI identifying cats in photographs in 72 hours of machine learning and after a cat is tagged in a few photographs (Cringely, 2013). AI could be applied to any field of study like finance, social science, science, engineering, etc. or even play against champions on the Jeopardy game show (Cyranoski, 2015; Cringely, 2013; Dewey, 2013; Power, 2015).

Example of artificial intelligence use in big data analysis: Genomics

The goal of AI use on genomic data is to help analyze physiological traits and lifestyle choices to provide a dedicated and personalized health plan to treat and eventually prevent disease (Cyranoski, 2015; Power, 2015).  This is done by feeding the AI systems with huge amounts of genomic data, which is considered big data by today’s standards (Cyranoski, 2015). Systems like IBM’s Watson (an AI system) could provide treatment options based on the results gained from analyzing thousands or even millions of genomic data (Power, 2015).  This is done by analyzing all this data and allowing machine learning techniques to devise algorithms based on the input data (Cringely, 2013; Cyranoski, 2015; Power, 2015).  As of 2015, there is about 100,000 individual genomic data in the system, and even with this huge amounts of data, it is still not enough to provide the personalized health plan that is currently being envisioned based on a person’s genomic data (Cyranoski, 2015).  Eventually, millions of individuals will need to be added into the AI system, and not just genomic data, but also proteomics, metabolomics, lipidomics, etc.

Resources:

Data tools: Analysis of big data involving text mining

Definitions

Big data – any set of data that has high velocity, volume, and variety, also known as the 3Vs (Davenport & Dyche, 2013; Fox & Do 2013, Podesta, Pritzker, Moniz, Holdren, & Zients, 2014).

Text mining – a process that involves discovering implicit knowledge from unstructured textual data (Gera & Goel, 2015; Hashimi & Hafez, 2015; Nassirtoussi Aghabozorgi, Wah, & Ngo, 2015).

Case study: Basole, Seuss, and Rouse (2013). IT innovation adoption by enterprises: Knowledge discovery through text analytics.

The goal of this study was to use text mining techniques on 472 quality peer reviewed articles that spanned 30 years of knowledge (1977-2008).  The selection criteria for the articles were based on articles focused on the adoption of IT innovation; focused on the enterprise, organization, or firm; rigorous research methods; and publishable leading journals.  The reason to go through all this analysis is to prove the usefulness of text analytics for literature reviews.  In 2016, most literature reviews contain recent literature from the last five years, and in certain fields, it may not just be useful to focus on the last five years.  Extending the literature search beyond this 5-year period, requires a ton of attention and manual labor, which makes the already literature an even more time-consuming endeavor than before. So, the author’s question is to see if it is possible to use text mining to conduct a more thorough review of the body of knowledge that expands beyond just the typical five years on any subject matter.  They argue that the time it takes to conduct this tedious task could benefit from automation.  However, this should be thought of as a first pass through the literature review. Thinking of this regarding a first pass allows for the generation of new research questions and a generation of ideas, which drives more future analysis.In the end, the study was able to conclude that cost and complexity were two of the most frequent determinants of IT innovation adoption from the perspective of an IT department.  Other determinants for IT departments were the complexity, capability, and relative advantage of the innovation.  However, when going up one level of extraction to the enterprise/organizational level, the perceived benefits and usefulness were the main determinants of IT innovation.  Ease of use of the technology was a big deal for the organization.  When comparing, IT innovation with costs there was a negative correlation between the two, while IT innovation has a positive correlation to organization size and top management support.

How was big data analytics learned, taught, and used in the case study?

The research approach for this study was: (1) Document Identification and extraction, (2) document classification and coding, (3) document analysis and knowledge discovery (key terms, co-occurrence), and (4) research gap identification.

Analysis of the data consisted of classifying the data into four time periods (bins): 1988-1979; 1980-1989; 1990-1999; and 2000-2008 and use of a classification scheme based on existing taxonomies (case study, content analysis, field experiment, field study, frameworks and conceptual model, interview, laboratory experiment, literature analysis, mathematical model, qualitative research, secondary data, speculation/commentary, and survey).  Data was also classified by their functional discipline (Information systems and computer science, decision science, management and organization sciences, economics, and innovation) and finally by IT innovation (software, hardware, networking infrastructure, and the tool’s IT term catalog). This study used a tool called Northernlight (http://georgiatech.northernlight.com/).

The hopes of this study are to use the bag-of-words technique and word proximity to other words (or their equivalents) to help extract meaning from a large set of text-based documents.  Bag-of-words technique is known for counting and identifying key terms and phrases, which help uncover themes.  The simplest way of thinking of the bag-of-words technique is word frequencies in a document.

However, understanding the meaning behind the themes means studying the context in which the words are located in, and relating them amongst other themes, also called co-occurrence of terms.  The best way of doing this meaning extraction is to measure the strength/distance between the themes.  Finally, the researcher in this study can set minimums, maximums that can enhance the meaning extraction algorithm to garner insights into IT innovation, while reducing the overall noise in the final results. The researchers set the following rules for co-occurrences between themes:

  • There are approximately 40 words per sentence
  • There are approximately 150 words per paragraph

How could this implementation of big data have been improved upon?

Goldbloom (2016) stated that using big data techniques (machine learning) is best on big data that requires classifying and it breaks down when the task is too small and specialized, therefore prime for only human analysis.  This study only looked at 427 articles, is this considered big enough for analysis, or should the analysis go back through multiple years beyond just the 30 years (Basole et al., 2013).  What is considered big data in 2013 (the time of this study), may not be big data in 2023 (Fox & Do, 2013).

Mei & Zhai (2005), observed how terms and term frequencies evolved over time and graphed it by year, rather than binning the data into four different groups as in Basole et al. (2013).  This case study could have shown how cost and complexity in IT innovation changed over time.  Graphing the results similar to Mei & Zhai (2005) and Yoon and Song (2014) would also allow for an analysis of IT innovation themes and if each of these themes is in an Introduction, Growth, Majority, or Decline mode.

 Reference

  • Basole, R. C., Seuss, C. D., & Rouse, W. B. (2013). IT innovation adoption by enterprises: Knowledge discovery through text analytics. Decision Support Systems, 54, 1044-1054. Retrieved from http://www.sciencedirect.com.ctu.idm.oclc.org/science/article/pii/S0167923612002849
  • Davenport, T. H., & Dyche, J. (2013). Big Data in Big Companies. International Institute for Analytics, (May), 1–31.
  • Fox, S., & Do, T. (2013). Getting real about Big Data: applying critical realism to analyse Big Data hype. International Journal of Managing Projects in Business, 6(4), 739–760. http://doi.org/10.1108/IJMPB-08-2012-0049
  • Gera, M., & Goel, S. (2015). Data Mining-Techniques, Methods and Algorithms: A Review on Tools and their Validity. International Journal of Computer Applications, 113(18), 22–29.
  • Goldbloom, A. (2016). The jobs we’ll lose to machines –and the ones we won’t. TED. Retrieved from http://www.ted.com/talks/anthony_goldbloom_the_jobs_we_ll_lose_to_machines_and_the_ones_we_won_t
  • Hashimi, H., & Hafez, A. (2015). Selection criteria for text mining approaches. Computers in Human Behavior, 51, 729–733. http://doi.org/10.1016/j.chb.2014.10.062
  • Mei, Q., & Zhai, C. (2005). Discovering evolutionary theme patterns from text: an exploration of temporal text mining. Proceedings of the Eleventh ACM SIGKDD International Conference on Knowledge Discovery in Data Mining, 198–207. http://doi.org/10.1145/1081870.1081895
  • Nassirtoussi, A. K., Aghabozorgi, S., Wah, T. Y., & Ngo, D. C. L. (2015). Text-mining of news-headlines for FOREX market prediction: a multi-layer dimension reduction algorithm with semantics and sentiment. Expert Systems with Applications, 42(1), 306-324.
  • Podesta, J., Pritzker, P., Moniz, E. J., Holdren, J., & Zients, J. (2014). Big Data: Seizing Opportunities. Executive Office of the President of USA, 1–79.
  • Yoon, B., & Song, B. (2014). A systematic approach of partner selection for open innovation. Industrial Management & Data Systems, 114(7), 1068.

Adv Quant: Compelling Topics

Compelling topics summary/definitions

  • Supervised machine learning algorithms: is a model that needs training and testing data set. However it does need to validate its model on some predetermined output value (Ahlemeyer-Stubbe & Coleman, 2014, Conolly & Begg, 2014).
  • Unsupervised machine learning algorithms: is a model that needs training and testing data set, but unlike supervised learning, it doesn’t need to validate its model on some predetermined output value (Ahlemeyer-Stubbe & Coleman, 2014, Conolly & Begg, 2014). Therefore, unsupervised learning tries to find the natural relationships in the input data (Ahlemeyer-Stubbe & Coleman, 2014).
  • General Least Squares Model (GLM): is the line of best fit, for linear regressions modeling along with its corresponding correlations (Smith, 2015). There are five assumptions to a linear regression model: additivity, linearity, independent errors, homoscedasticity, and normally distributed errors.
  • Overfitting: is stuffing a regression model with so many variables that have little contributional weight to help predict the dependent variable (Field, 2013; Vandekerckhove, Matzke, & Wagenmakers, 2014). Thus, to avoid the over-fitting problem, the use of parsimony is important in big data analytics.
  • Parsimony: is describing a dependent variable with the fewest independent variables as possible (Field, 2013; Huck, 2013; Smith, 2015). The concept of parsimony could be attributed to Occam’s Razor, which states “plurality out never be posited without necessity” (Duignan, 2015).  Vandekerckhove et al. (2014) describe parsimony as a way of removing the noise from the signal to create better predictive regression models.
  • Hierarchical Regression: When the researcher builds a multivariate regression model, they build it in stages, as they tend to add known independent variables first, and add newer independent variables in order to avoid overfitting in a technique called hierarchical regression (Austin, Goel & van Walraven, 2001; Field, 2013; Huck 2013).
  • Logistic Regression: multi-variable regression, where one or more independent variables are continuous or categorical which are used to predict a dichotomous/ binary/ categorical dependent variable (Ahlemeyer-Stubbe, & Coleman, 2014; Field, 2013; Gall, Gall, & Borg, 2006; Huck, 2011).
  • Nearest Neighbor Methods: K-nearest neighbor (i.e. K =5) is when a data point is clustered into a group, by having 5 of the nearest neighbors vote on that data point, and it is particularly useful if the data is a binary or categorical (Berson, Smith, & Thearling, 1999).
  • Classification Trees: aid in data abstraction and finding patterns in an intuitive way (Ahlemeyer-Stubbe & Coleman, 2014; Brookshear & Brylow, 2014; Conolly & Begg, 2014) and aid the decision-making process by mapping out all the paths, solutions, or options available to the decision maker to decide upon.
  • Bayesian Analysis: can be reduced to a conditional probability that aims to take into account prior knowledge, but updates itself when new data becomes available (Hubbard, 2010; Smith, 2015; Spiegelhalter & Rice, 2009; Yudkowsky, 2003).
  • Discriminate Analysis: how should data be best separated into several groups based on several independent variables that create the largest separation of the prediction (Ahlemeyer-Stubbe, & Coleman, 2014; Field, 2013).
  • Ensemble Models: can perform better than a single classifier, since they are created as a combination of classifiers that have a weight attached to them to properly classify new data points (Bauer & Kohavi, 1999; Dietterich, 2000), through techniques like Bagging and Boosting. Boosting procedures help reduce both bias and variance of the different methods, and bagging procedures reduce just the variance of the different methods (Bauer & Kohavi, 1999; Liaw & Wiener, 2002).

 

References

  • Ahlemeyer-Stubbe, Andrea, Shirley Coleman. (2014). A Practical Guide to Data Mining for Business and Industry, 1st Edition. [VitalSource Bookshelf Online].
  • Austin, P. C., Goel, V., & van Walraven, C. (2001). An introduction to multilevel regression models. Canadian Journal of Public Health, 92(2), 150.
  • Bauer, E., & Kohavi, R. (1999). An empirical comparison of voting classification algorithms: Bagging, boosting, and variants. Machine learning,36(1-2), 105-139.
  • Berson, A. Smith, S. & Thearling K. (1999). Building Data Mining Applications for CRM. McGraw-Hill. Retrieved from http://www.thearling.com/text/dmtechniques/dmtechniques.htm
  • Brookshear, G., & Brylow, D. (2014). Computer Science: An Overview, 12th Edition. [VitalSource Bookshelf Online].
  • Connolly, T., & Begg, C. (2014). Database Systems: A Practical Approach to Design, Implementation, and Management, 6th Edition. [VitalSource Bookshelf Online].
  • Dietterich, T. G. (2000). Ensemble methods in machine learning. International workshop on multiple classifier systems (pp. 1-15). Springer Berlin Heidelberg.
  • Duignan, B. (2015). Occam’s razor. Encyclopaedia Britannica. Retrieved from https://www.britannica.com/topic/Occams-razor
  • Field, Andy. (2013). Discovering Statistics Using IBM SPSS Statistics, 4th Edition. [VitalSource Bookshelf Online].
  • Gall, M. D., Gall, J. P., Borg, W. R. (2006). Educational Research: An Introduction, 8th Edition. [VitalSource Bookshelf Online].
  • Hubbard, D. W. (2010). How to measure anything: Finding the values of “intangibles” in business. (2nd e.d.) New Jersey, John Wiley & Sons, Inc.
  • Huck, Schuyler W. (2011). Reading Statistics and Research, 6th Edition. [VitalSource Bookshelf Online].
  • Liaw, A., & Wiener, M. (2002). Classification and regression by randomForest. R news, 2(3), 18-22.
  • Smith, M. (2015). Statistical analysis handbook. Retrieved from http://www.statsref.com/HTML/index.html?introduction.html
  • Spiegelhalter, D. & Rice, K. (2009) Bayesian statistics. Retrieved from http://www.scholarpedia.org/article/Bayesian_statistics
  • Vandekerckhove, J., Matzke, D., & Wagenmakers, E. J. (2014). Model comparison and the principle of parsimony.
  • Yudkowsky, E.S. (2003). An intuitive explanation of Bayesian reasoning. Retrieved from http://yudkowsky.net/rational/bayes