Professional
decision making often takes place just about anywhere: at the
office, walking through the park, talking with the customer, while
drinking coffee, etc. To support and raise the quality of
decisions, computerized decision support tools are being
constructed for various purposes. Their major advantage is their
speed, support of systematic approach to decision making, ability
to explain decisions, and objectively - according to decision mode
- rank (possibly a large set of) variants.
Their biggest disadvantage, though, is the computer, which often
is not available where the decisions are made. In medicine, for
instance, a physician can not take a break with his patient to jump
into his office, switch on the computer, run the software, type in
the data, and then run back to the patient (which is right then
probably knocking at the office of a colleague-physician).
We are building a set of decision support shells that should work both on handheld computers and on web to alleviate these problems. We call
our schema decisions-at-hand. In this schema, machine learning (data
mining) and/or statistical analysis is used to develop decision
models from data; these are then encoded in XML, and
used in decision shells either on web or on a handheld device.
When fully developed, the decisions-at-hand will consists of:
Web-based decision support shell. The software currently supports logistic regression and naive bayesian decision support models, other models may be encoded through feature transformations.
PocketPC-based decision support shell (coming in March 2003). Early implementations were based on PalmOS.
Wizards that, given the data set, construct predictive models and encode them in XML. For this, we will use Orange and Orange Widgets platforms. A web-wizard for construction of naive Bayesian model will be availbe in May 2003.
Decisions-at-hand schema is being tested on several applications,
including:
Prostate
Cancer Prognostics (with Mike Kattan from Memorial Sloan-Kettering
Cancer Center, New York, NY)
Prognostics
in Trauma and Emergency Medicine (with Noriaki Aoki and J. Robert
Beck, Baylor College of Medicine, Houston, TX)
Prognostics
and Therapy evaluation after severe hip and hand injuries (with
Dragica Smrke, Clinical Center, Ljubljana, Slovenia)