Limits...
Personalized Modeling for Prediction with Decision-Path Models.

Visweswaran S, Ferreira A, Ribeiro GA, Oliveira AC, Cooper GF - PLoS ONE (2015)

Bottom Line: Deriving predictive models in medicine typically relies on a population approach where a single model is developed from a dataset of individuals.We introduce three personalized methods that derive personalized decision-path models.Two of the three personalized methods performed statistically significantly better on area under the ROC curve (AUC) and Brier skill score compared to CART.

View Article: PubMed Central - PubMed

Affiliation: Department of Biomedical Informatics, University of Pittsburgh, Pittsburgh, Pennsylvania, United States of America; The Intelligent Systems Program, University of Pittsburgh, Pittsburgh, Pennsylvania, United States of America.

ABSTRACT
Deriving predictive models in medicine typically relies on a population approach where a single model is developed from a dataset of individuals. In this paper we describe and evaluate a personalized approach in which we construct a new type of decision tree model called decision-path model that takes advantage of the particular features of a given person of interest. We introduce three personalized methods that derive personalized decision-path models. We compared the performance of these methods to that of Classification And Regression Tree (CART) that is a population decision tree to predict seven different outcomes in five medical datasets. Two of the three personalized methods performed statistically significantly better on area under the ROC curve (AUC) and Brier skill score compared to CART. The personalized approach of learning decision path models is a new approach for predictive modeling that can perform better than a population approach.

No MeSH data available.


Related in: MedlinePlus

Pseudocode for the DP-AUC method.
© Copyright Policy
Related In: Results  -  Collection

License
getmorefigures.php?uid=PMC4476684&req=5

pone.0131022.g003: Pseudocode for the DP-AUC method.

Mentions: The pseudocode for DP-AUC method is given in Fig 3. DP-AUC differs from the other two algorithms in that it uses AUC to evaluate S’ which is obtained from path S by temporarily adding a candidate variable V. Computing the AUC requires obtaining a prediction for T for each person in the training set of individuals that matches S’. This is done using leave-one-out cross validation. Specifically, one of the training individuals that matches S’ is left out and Eq (1) is used to estimate the parameters of the distribution over T for the left-out person from the remaining training individuals that match S’. Using the probability parameters, a probabilistic prediction is obtained for T for the left-out person with the following equation:P(T=p/VS=vS)=θp,(7)where p denotes the positive value of the target variable and θp is the parameter estimate for the positive value. The process is repeated by leaving out each of the training individuals in turn so that a prediction is obtained for each person. Using this leave-on-out method for computing predictions, the AUC for S’ is computed from the predictions for the individuals in DTemp and the individuals in D–DTemp (see line 11 in Fig 3).


Personalized Modeling for Prediction with Decision-Path Models.

Visweswaran S, Ferreira A, Ribeiro GA, Oliveira AC, Cooper GF - PLoS ONE (2015)

Pseudocode for the DP-AUC method.
© Copyright Policy
Related In: Results  -  Collection

License
Show All Figures
getmorefigures.php?uid=PMC4476684&req=5

pone.0131022.g003: Pseudocode for the DP-AUC method.
Mentions: The pseudocode for DP-AUC method is given in Fig 3. DP-AUC differs from the other two algorithms in that it uses AUC to evaluate S’ which is obtained from path S by temporarily adding a candidate variable V. Computing the AUC requires obtaining a prediction for T for each person in the training set of individuals that matches S’. This is done using leave-one-out cross validation. Specifically, one of the training individuals that matches S’ is left out and Eq (1) is used to estimate the parameters of the distribution over T for the left-out person from the remaining training individuals that match S’. Using the probability parameters, a probabilistic prediction is obtained for T for the left-out person with the following equation:P(T=p/VS=vS)=θp,(7)where p denotes the positive value of the target variable and θp is the parameter estimate for the positive value. The process is repeated by leaving out each of the training individuals in turn so that a prediction is obtained for each person. Using this leave-on-out method for computing predictions, the AUC for S’ is computed from the predictions for the individuals in DTemp and the individuals in D–DTemp (see line 11 in Fig 3).

Bottom Line: Deriving predictive models in medicine typically relies on a population approach where a single model is developed from a dataset of individuals.We introduce three personalized methods that derive personalized decision-path models.Two of the three personalized methods performed statistically significantly better on area under the ROC curve (AUC) and Brier skill score compared to CART.

View Article: PubMed Central - PubMed

Affiliation: Department of Biomedical Informatics, University of Pittsburgh, Pittsburgh, Pennsylvania, United States of America; The Intelligent Systems Program, University of Pittsburgh, Pittsburgh, Pennsylvania, United States of America.

ABSTRACT
Deriving predictive models in medicine typically relies on a population approach where a single model is developed from a dataset of individuals. In this paper we describe and evaluate a personalized approach in which we construct a new type of decision tree model called decision-path model that takes advantage of the particular features of a given person of interest. We introduce three personalized methods that derive personalized decision-path models. We compared the performance of these methods to that of Classification And Regression Tree (CART) that is a population decision tree to predict seven different outcomes in five medical datasets. Two of the three personalized methods performed statistically significantly better on area under the ROC curve (AUC) and Brier skill score compared to CART. The personalized approach of learning decision path models is a new approach for predictive modeling that can perform better than a population approach.

No MeSH data available.


Related in: MedlinePlus