Many machine learning algorithms require a single value per feature per record for modeling. However, there are applications, in the medical domain particularly, where a single record may have multiple observations for the same feature. For example, a patient could have the same gene analyzed in multiple tissue slides of a biopsy, or could have the same genetic test performed on multiple subsequent biopsies. The challenge in these applications is how to integrate multiple observations of the same predictor feature per record. In this paper, two data aggregation methods are compared, one method is a simple median aggregation of feature values, while the other is a novel method which constructs intervals of values for each feature. The aggregated features are passed as input to a novel support vector regression method for modeling survival data in a prostate cancer setting. The performance of both methods was similar in predicting prostate cancer progression on three data cohorts.

A crucial challenge in predictive modeling for survival analysis is managing censored observations in the data. The Cox proportional hazards model is the standard tool for the analysis of continuous censored survival data. We propose a novel machine learning algorithm, support vector regression for censored data (SVRc) for improved analysis of medical survival data. SVRc leverages the high-dimensional capabilities of traditional SVR while adapting it for use with censored data through a modified asymmetric loss/penalty function which allows censored (left and right censored) data to be processed. We applied the new algorithm to predict the recurrence and disease progression of prostate cancer, breast cancer and lung cancer. Compared with the traditional Cox model, SVRc achieves significant improvement in overall accuracy as well as in the ability to identify high-risk and low-risk patient populations.

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, which is the sum of measurement costs and misdiagnosis costs. In most diagnostic settings, there is a tradeoff between these two kinds of costs. This paper formalizes diagnostic decision making as a Markov Decision Process (MDP). The paper introduces a new family of systematic search algorithms based on the AO* algorithm to solve this MDP. 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. To reduce overfitting, regularizers are integrated into the search algorithms. 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.