Statement of Problem for Advanced Clinical Decision Making
Using disparate data sources streaming across all Intensive Care Units within a major hospital setting, several key issues are addressed in this chapter. Clinically we show a method to improve the current one dimensional high-level alarms exhibited within current practices, that needlessly distract bedside care providers away from their patients, by creating a multivariate composite risk score of mortality. Statistically we show how to transform streaming information and take advantage of the time dimension provided by the transactions timestamp. From a metadata perspective we show how to JOIN these different timed-stamped transactions into a common time-dimensioned packet of information. From a systems resource demand on data storage, we show how to use an optimized binning method to represent only those critical ranges within each variables distribution. These advanced computational techniques should be useful in generating a higher-level picture of the patient that can support more effective clinical decision support.
Specific Aim 1 : Expand upon a variety of data reduction techniques that allow researchers to distill vast amounts of disparate streaming data sources by quantitatively binning and tracking shifts across time.
Specific Aim 2 : Interface these multiple disparate forms of clinically important patient data into a common time framed, multivariate packet of information.
Specific Aim 3 : Develop Mortality predictions based on statistical principles, but utilizing novel neural network techniques and other non-linear trending techniques.
Specific Aim 4 : Develop Novel graphical interface that allows bedside care givers to understand and act upon large sets of clinically important data
Statement of Problem for Fundamental Equities Selection
Using data from Compustat to predict percent and absolute dollar gainers several key issues are addressed in this chapter. From a portfolio management perspective we show a method to improve upon using raw values to model with and that vary in meaning quarter to quarter, by creating a relative ranking transformation such that top market performers are always coded together. Statistically we show how to time-shift n-periods ahead and how percent gainers differ from absolute dollar gainers. From a metadata perspective we show how to combine individual financial ratio-ranks into combinatoric sequences with significantly higher lift from base. From a systems resource demand on data storage, we show how to use only those sequences to represent those critical rules to optimize the portfolios metrics. These advanced computational techniques should be useful in generating a highly selected portfolio management rules that can support more consistent holding period return.
But how is Multivariate different..REALLY !
In this highly competitive environment I need to point out some differences that are really critical. I am focused on timestamped transactions from both WEBLOGS and OFFLINE Master files. For a Mortgage Originators' inbound call center where TV ADS and WEB paid and natural keywords drive all activity while under dramatic media budget cuts. For a Pharmaceutical companys' unbranded website gathering pre-qualified emails from patients, separating caregivers and professionals. For an entertainment star online retailing of songs, ring tones, perfume, and accessories. For a Liquor distributor holding 1,000 tasting events a week, where site profiles impact ROI. For RMBS Bond dealer that is quantitatively warehousing and pricing new deals using a bottoms up loan level methodology. For a 24 hour store selling shelf space based upon POS sales. For a website developer that confuses visitors with a bad landing page. For a CPC site that is losing money with huge Bounce Back Rates.
Common among these efforts is dynamic streaming data that is mapped with other disparate informations sources, then transformed, modeled, and dashboarded into ACTIONABLE Business Intelligence.
Yes, I use a BIG toolkit of everything expected but I am not selling SAS, SQL, COGNOS, ORACLE expertise per diem, but providing multivariate solutions based upon your data which are generally NOT apparent to you.
Waterfall analysis identifies the fall off turning initial transactions into sales. Who cares how many impressions lead to how many visitors, when its the email address or paid download that counts. Vintage analysis takes snap shot online reports (Omniture, Webtrends, etc) and JOINS them into your companies master files which expects future sales and ties ALL activity of that customer together for a TRUE 360 View.
|
|