Robert Nisbet

Text Mining and Natural Language Processing

Pattern recognition is the most basic description of what is done in the process of data mining. Text mining is the process of deriving novel information from a collection of texts. It can be applied to many applications in a variety of fields, namely, marketing, national security, medical and biomedical, and public relation. The process of counting the number of matches to a text pattern occurs repeatedly in text mining, such that one can compare two different documents by counting how many times different words occur in each document. Analysts choose the best way to analyze the text further, by either combining groups of words that appear to mean the same thing or directing the computer to do it automatically in a second iteration of the process, and then analyze the results. The goals of text mining include identification of sets of related words in documents, identification of clusters of similar reports, exploratory analysis using structured (fields in a record) and unstructured data (textual information) to discover hidden patterns that could provide some useful insights related to causes of fatal accidents, and identification of frequent item sets. The Feature Selection and Variable Screening tool is extremely useful for reducing the dimensionality of analytic problems.

IBM Watson for Clinical Decision Support

One approach to the design of a medical decisioning system is to avoid making specific decisions, but rather scan through the hugely voluminous structured (databases) and unstructured (text) data sources, and present a list of evidence-based alternatives. These alternatives can be submitted to the most sophisticated non-linear analytical processing system in the universe: the minds of the physicians charged with providing the right diagnosis and treatment for the right patient at the right time. This chapter describes an early attempt to use the IBM Watson computer (winner of the TV contest show, Jeopardy!) to create such a tool, which in the hands of the physician is a very sophisticated extension of his or her own brain.

Basic Algorithms for Data Mining: A Brief Overview

This chapter discusses the basic algorithms used in data mining and helps to select the right one to use. It presents two semi-automated approaches to performing all the necessary operations from accessing data to producing model results. The first example shows how STATISTICA Data Miner Recipe (DMR recipe) Interface packages all basic steps of a data mining project into an easy-to-use interface. The second example is KXEN (Knowledge Extraction Engine). Both tools select the modeling algorithms and permit to enter a few settings, and automatically generate model results. Use of either tool might be the best way for beginning data miners to build their first model. The DMRecipe Interface provides a step-by-step approach to data preparation, variable selection, and dimensionality reduction, resulting in models trained with different algorithms. The automated functions of DMRecipe and KXEN Modeling Assistant provide a glimpse of one direction in which data mining is developing. These tools provide a close analogy in which data mining is as easy to use as the automobile interface. Association algorithms can be used to analyze simple categorical variables, dichotomous variables, and/or multiple target variables. The goal of association rules is to detect relationships or associations between specific values of categorical variables in large data sets. This technique allows analysts and researchers to uncover hidden patterns in large data sets.

The Data Mining Process

Data miners state that data mining is as much art as it is science. The concept of data mining to a business data analyst includes not only the finding of relationships, but also the necessary preprocessing of data, interpretation of results, and provision of the mined information in a form useful in decision-making. The method followed in the data mining process for business is a blend of the mathematical and scientific methods. The basic data mining process flow follows the mathematical method, but some steps from the scientific method are included. The business objective is to find a way to capture relevant information in these unstructured formats into a data format that will support decision making. Evaluation of modeling results should include a list of possible modeling goals for the future and the modeling approaches to accomplish them. The modeling report should discuss briefly the steps involved and how to accomplish them. These steps should be expressed in terms of what support must be gained among the stakeholders targeted by these new projects, the processes in the company that must be put in place to accomplish the new projects, and the expected benefit to the company for doing so.

Chapter 3 – Biomedical Informatics

This chapter is an introduction to proactive decisioning in medicine and health care, facilitated by the construction of analytical models to predict future states, rather than react to existing healthcare conditions.

Platform for Data Integration and Analysis, and Publishing Medical Knowledge as Done in a Large Hospital

The Military Institute of Medicine in Poland had a problem. They wanted to build a predictive analytics decisioning system, but realized that it had to be designed as a whole system – they could not just cobble together some existing parts and resources. They designed the entire system from scratch, including standardized data access from a variety of sources, data preparation, and data analysis to guide medical diagnosis and treatment. This chapter describes how they did it.

The Nature of Insight from Data and Implications for Automated Decisioning: Predictive and Prescriptive Models, Decisions, and Actions

In this chapter we move into the next phase of the predictive analytics operation – combining the various hardware and software elements into an automated decision-making system. These “decisioning” systems are the goal; the “Holy Grail” of predictive analytics. Individual algorithms and methodologies should not be viewed as ends in themselves, but as means to an end – the predictive analytics systems that receive input data, conduct necessary data preparation operations, train the appropriate modeling algorithms, and output the decisions themselves – not just some information that can be used by subjective humans to make decisions

Root Cause Analysis, Six Sigma, and Overall Quality Control and Lean Concepts: The First Process to Bring Quality and Cost-Effectiveness to Medical Care Delivery

In 1996, the Institute of Medicine (IOM) launched a program to improve the quality of health in the nation. This focus led to the incorporation of the Six Sigma predictive analytical process developed in other industries. The combination of Six Sigma processing and Deming’s emphasis on continuous quality improvement led to the development of the fishbone process model. The core element of the fishbone model is root cause analysis. This chapter presents a rich landscape of quality control issues in health care, with a focus on root cause analysis in hospitals.

Predictive Analytics in Nursing Informatics

From the beginning of medical treatment in our civilization, the prevailing view was that medicine was a “noble” art practiced by the physician, rather than a science pursued by the learned. That view began to change in the mid-1800s, instigated not by a physician but by a nurse – Florence Nightingale. The intention of this chapter is to inspire the nurse informatician with the possibilities of predictive analytics in nursing informatics, and examples of current research projects. There is endless potential in this field for effectively using predictive analytics and data mining. This chapter is not meant to be a complete assessment of the possibilities, or to describe all the areas in which PA can be used, but rather to be a guide with examples to stimulate the mind. Modern nursing informatics, as presented in this chapter, follows in Florence Nightingale’s footsteps.

Model Evaluation and Enhancement

Publisher Summary This chapter discusses ways to explain how well the model is doing and then gives a checklist of actions one can employ to improve its performance. Using a reliable technique for model assessment is essential. The essential first step in any modeling task is to split off an evaluation set. Statisticians have long known of this relationship between complexity and accuracy, and one way to avoid overfit is to regulate the complexity of the model. Methods traditionally used in statistical analysis often contribute significantly to a data mining effort, at the very least providing a baseline against which to compare more modern techniques. Linear discriminant analysis (LDA) predicts a categorical response variable by creating a discriminating plane separating the groups of the response variable. A quadratic extension allows for nonlinear boundaries but requires estimating covariance matrices for each class. Cluster analysis divides a heterogeneous group of records into several more homogeneous classes, or clusters. These clusters contain records that are similar in their values on particular variables. Many algorithms prefer the variables to be on the same scale and be independent.