Technical Report


Integrating learning from examples into the search for diagnostic policies Public Deposited

Downloadable Content

Download PDF


Attribute NameValues
  • This paper studies the problem of learning diagnostic policies from training examples. A diagnostic policy is a complete description of the decision-making actions of a diagnostician (i.e., tests followed by a diagnostic decision) for all possible combinations of test results. An optimal diagnostic policy is one that minimizes the expected total cost of diagnosing a patient, where the cost is the sum of two components: (a) measurement costs (the costs of performing various diagnostic tests) and (b) misdiagnosis costs (the costs incurred when the patient is incorrectly diagnosed). In most diagnostic settings, there is a tradeoff between these two kinds of costs. A diagnostic policy that minimizes measurement costs usually performs fewer tests and tends to make more diagnostic errors, which are expensive. Conversely, a policy that minimizes misdiagnosis costs usually makes more measurements. This paper formalizes diagnostic decision making as a Markov Decision Process (MDP). It then presents a range of algorithms for solving this MDP. These algorithms can be divided into methods based on systematic search and methods based on greedy search. The paper introduces a new family of systematic algorithms based on the AO* algorithm. To make AO* efficient, the paper describes an admissible heuristic that enables AO* to prune large parts of the search space. The paper also introduces several greedy algorithms including some improvements over previously-published methods. The paper then addresses the question of learning diagnostic policies from examples. When the probabilities of diseases and test results are computed from training data, there is a great danger of overfitting. The paper introduces a range of regularization methods to reduce overfitting. An interesting aspect of these regularizers is that they are integrated into the search algorithms rather than being isolated in a separate learning step prior to searching for a good diagnostic policy. Finally, the paper compares the proposed methods on five benchmark diagnostic data sets. The studies show that in most cases the systematic search methods produce better diagnostic policies than the greedy methods. In addition, the studies show that for training sets of realistic size, the systematic search algorithms are practical on today's desktop computers. Hence, these AO*-based methods are recommended for learning diagnostic policies that seek to minimize the expected total cost of diagnosis.
Resource Type
Date Available
Date Issued
Rights Statement
Peer Reviewed
Additional Information
  • description.provenance : Approved for entry into archive by Laura Wilson( on 2012-12-04T16:29:41Z (GMT) No. of bitstreams: 1 2004-33.pdf: 409050 bytes, checksum: b697e8620e349b83dbc236faf67def1d (MD5)
  • description.provenance : Made available in DSpace on 2012-12-04T16:29:41Z (GMT). No. of bitstreams: 1 2004-33.pdf: 409050 bytes, checksum: b697e8620e349b83dbc236faf67def1d (MD5) Previous issue date: 2004
  • description.provenance : Submitted by Laura Wilson ( on 2012-12-04T16:28:42Z No. of bitstreams: 1 2004-33.pdf: 409050 bytes, checksum: b697e8620e349b83dbc236faf67def1d (MD5)



This work has no parents.