#### DMCA

## Statistical relational learning to predict primary myocardial infarction from electronic health records (2012)

Venue: | AAAI Conference on Innovative Applications in AI (IAAI |

Citations: | 2 - 2 self |

### Citations

995 | Greedy function approximation: a gradient boosting machine - Friedman - 2001 |

436 | Policy gradient methods for reinforcement learning with function approximation
- Sutton, McAllester, et al.
- 2000
(Show Context)
Citation Context ...ng examples. So ascent in the direction of hm will approximate the true functional gradient. The same idea has later been used to learn several relational models and policies [Natarajan et al., 2010; =-=Sutton et al., 2000-=-; Kersting and Driessens, 2008; Natarajan et al., 2011; Gutmann and Kersting, 2006]. Let us denote the MI as y and it is binary valued (i.e., occurrence of MI). Let us denote all the other variables m... |

337 |
Introduction to statistical relational learning
- Getoor, Taskar
- 2007
(Show Context)
Citation Context ...n of high-risk subjects, which a predictive model can accurately identify. The primary approach we use draws from relational probabilistic models, also known as Statistical Relational Learning (SRL) [=-=Getoor and Taskar, 2007-=-]. Their primary advantage is their ability to work with the structure and relations in data; that is, information about one object helps the learning algorithms to reach conclusions about other objec... |

143 | Learning relational probability trees
- Neville, Jensen, et al.
- 2003
(Show Context)
Citation Context ... other objects. Unfortunately, most SRL algorithms have difficulty scaling to large data sets. One efficient approach that yields good results from large data sets is the relational probability tree [=-=Neville et al., 2003-=-]. The performance increase observed moving from propositional decision trees to forests is also seen in the relational domain [Anderson and Pfahringer, 2009; Natarajan et al., 2010]. One method calle... |

117 | Extracting tree-structured representations of trained networks.
- Craven, Shavlik
- 1996
(Show Context)
Citation Context ..., as frequent lipoprotein measurements may display a concern for atherosclerosis-related illness. The set of trees can also be converted into a list of weighted rules to make them more interpretable [=-=Craven and Shavlik, 1996-=-]. The density plot in Figure 5 shows the ability of RFGB and RPT models to separate the MI class from the controls. It is clear from the far left region of the RFGB graph that we can accurately ident... |

111 |
Top-down induction of first-order logical decision trees
- Blockeel, Raedt
- 1998
(Show Context)
Citation Context ...demonstrated to build significantly smaller trees than other conditional models and obtain comparable performance. We use a version of RPTs that employs the TILDE relational regression (RRT) learner [=-=Blockeel and Raedt, 1998-=-] where we learn a regression tree to predict positive examples (in this case, patients with MI) and turn the regression values in the leaves into probabilities by exponentiating the regression value ... |

78 | Training conditional random fields via gradient tree boosting - Dietterich, Ashenfelter, et al. - 2004 |

74 |
ACCF/AHA Guideline for Assessment of Cardiovascular Risk in Asymptomatic Adults:
- Greenland, JS, et al.
- 2010
(Show Context)
Citation Context ...ow-density lipoprotein (LDL) cholesterol, diabetes, obesity, inactivity, alcohol and smoking. Studies have also identified less common risk factors as well as subgroups with particular risk profiles [=-=Greenland et al., 2010-=-; Antonopoulos, 2002]. The canonical method of study in this field is the identification or quantification of the risk attributable to a variable in isolation using: case-control studies, cohort studi... |

37 | TildeCRF: Conditional random fields for logical sequences
- Gutmann, Kersting
- 2006
(Show Context)
Citation Context ...ng to each path from the root to a leaf. The resulting potential functions from all these different RRTs still have the form of a linear combination of features but the features can be quite complex [=-=Gutmann and Kersting, 2006-=-]. We use weighted variance as the criterion to split on in the inner nodes. We augment the RRT learner with aggregation functions such as count, max, average that are used in the standard SRL literat... |

28 | Non–parametric policy gradients: A unified treatment of propositional and relational domains
- Kersting, Driessens
- 2008
(Show Context)
Citation Context ...t in the direction of hm will approximate the true functional gradient. The same idea has later been used to learn several relational models and policies [Natarajan et al., 2010; Sutton et al., 2000; =-=Kersting and Driessens, 2008-=-; Natarajan et al., 2011; Gutmann and Kersting, 2006]. Let us denote the MI as y and it is binary valued (i.e., occurrence of MI). Let us denote all the other variables measured over the different yea... |

16 | Imitation learning in relational domains: A functionalgradient boosting approach
- Natarajan, Joshi, et al.
- 2011
(Show Context)
Citation Context ...approximate the true functional gradient. The same idea has later been used to learn several relational models and policies [Natarajan et al., 2010; Sutton et al., 2000; Kersting and Driessens, 2008; =-=Natarajan et al., 2011-=-; Gutmann and Kersting, 2006]. Let us denote the MI as y and it is binary valued (i.e., occurrence of MI). Let us denote all the other variables measured over the different years as x. Hence, we are i... |

13 |
Marshfield Clinic personalized medicine research project (PMRP): design, methods, and recruitment for a large population-based biobank. Personalized Medicine 2005;2(1):49–79
- McCarty, Wilke, et al.
(Show Context)
Citation Context ...Flow chart depicting experimental setup Experimental Methods We analyzed de-identified EHR data on 18, 386 subjects enrolled in the Personalized Medicine Research Project (PMRP) at Marshfield Clinic [=-=McCarty et al., 2005-=-; 2008]. The PMRP cohort is one of the largest population-based biobanks in the United States and consists of individuals who are 18 years of age or older, who have consented to the study and provided... |

12 | Third report of the National Cholesterol Education Program (NCEP) expert panel on detection, evaluation and treatment of high blood cholesterol in adults (Adult Treatment Panel III) final report. - Antonopoulos - 2002 |

5 |
The WEKA data mining software: an update. Special Interest Group on Knowledge Discovery and Data Mining Explorer Newsletter
- Hall, Frank, et al.
- 2009
(Show Context)
Citation Context ... machines (SVMs; linear kernel, C 1.0; radial basis function kernel, C 250007, G 0.01), and random forests (RF; 10 trees, default parameters). All propositional learners were run using Weka software [=-=Hall et al., 2009-=-]. In our secondary analysis, we varied both the experimental setup and the RFGB parameters to investigate the effect on their predictive ability. First, we altered the case-control ratio {1:1, 1:2, 1... |

5 | Boosting relational dependency networks
- Natarajan, Khot, et al.
- 2010
(Show Context)
Citation Context ...al probability tree [Neville et al., 2003]. The performance increase observed moving from propositional decision trees to forests is also seen in the relational domain [Anderson and Pfahringer, 2009; =-=Natarajan et al., 2010-=-]. One method called functional gradient boosting (FGB) has achieved good performance in the propositional domain [Friedman, 2001]. We apply it to the relational domain for our task: the prediction an... |

2 | Relational Random Forests Based on Random Relational Rules - Anderson, Pfahringer - 2009 |

1 |
Predictors of acute myocardial infarction mortality in hypertensive patients treated in primary care. Scandinavian Journal of Primary Health Care 25(4):237–243
- Bg-Hansen, Larsson, et al.
- 2007
(Show Context)
Citation Context ... these studies, data is collected at the study onset t0 to serve as the baseline variables, whose values are the ones used to determined risk. To illustrate this, consider the Skaraborg cohort study [=-=Bg-Hansen et al., 2007-=-] for the identification of acute MI mortality risk factors. The study measured established risk factors for MI at t0, and then the subjects participated in annual checkups to assess patient health an... |

1 | 2002. Prediction of mortality from coronary heart disease among diverse populations: is there a common predictive function? Heart 88:222–228 - Group |