@Proceedings{CPC2008,
title = {Proceedings of Machine Learning Research},
booktitle = {Proceedings of Machine Learning Research},
editor = {Isabelle Guyon and Constantin Aliferis and Greg Cooper and André Elisseeff and Jean-Philippe Pellet and Peter Spirtes and Alexander Statnikov},
publisher = {PMLR},
series = {Proceedings of Machine Learning Research},
volume = 3
}
@InProceedings{guyon08a,
title = {Design and Analysis of the Causation and Prediction Challenge},
author = {Isabelle Guyon and Constantin Aliferis and Greg Cooper and André Elisseeff and Jean-Philippe Pellet and Peter Spirtes and Alexander Statnikov},
booktitle = {Proceedings of the Workshop on the Causation and Prediction Challenge at WCCI 2008},
pages = {1--33},
year = {2008},
editor = {Isabelle Guyon and Constantin Aliferis and Greg Cooper and André Elisseeff and Jean-Philippe Pellet and Peter Spirtes and Alexander Statnikov},
volume = {3},
series = {Proceedings of Machine Learning Research},
address = {Hong Kong},
month = {03--04 Jun},
publisher = {PMLR},
pdf = {http://proceedings.mlr.press/v3/guyon08a/guyon08a.pdf},
url = {http://proceedings.mlr.press/v3/guyon08a.html},
abstract = {We organized for WCCI 2008 a challenge to evaluate causal modeling techniques, focusing on predicting the effect of “interventions” performed by an external agent. Examples of that problem are found in the medical domain to predict the effect of a drug prior to administering it, or in econometrics to predict the effect of a new policy prior to issuing it. We concentrate on a given target variable to be predicted (e.g., health status of a patient) from a number of candidate predictive variables or “features” (e.g., risk factors in the medical domain). Under interventions, variable predictive power and causality are tied together. For instance, both smoking and coughing may be predictive of lung cancer (the target) in the absence of external intervention; however, prohibiting smoking (a possible cause) may prevent lung cancer, but administering a cough medicine to stop coughing (a possible consequence) would not. We propose four tasks from various application domains, each dataset including a training set drawn from a “natural” distribution and three test sets: one from the same distribution as the training set and two corresponding to data drawn when an external agent is manipulating certain variables. The goal is to predict a binary target variable, whose values on test data are withheld. The participants were asked to provide predictions of the target variable on test data and the list of variables (features) used to make predictions. The challenge platform remains open for post-challenge submissions and the organization of other events is under way (see http://clopinet.com/causality).}
}
@InProceedings{brown08a,
title = {A Strategy for Making Predictions Under Manipulation},
author = {Laura E. Brown and Ioannis Tsamardinos},
booktitle = {Proceedings of the Workshop on the Causation and Prediction Challenge at WCCI 2008},
pages = {35--52},
year = {2008},
editor = {Isabelle Guyon and Constantin Aliferis and Greg Cooper and André Elisseeff and Jean-Philippe Pellet and Peter Spirtes and Alexander Statnikov},
volume = {3},
series = {Proceedings of Machine Learning Research},
address = {Hong Kong},
month = {03--04 Jun},
publisher = {PMLR},
pdf = {http://proceedings.mlr.press/v3/brown08a/brown08a.pdf},
url = {http://proceedings.mlr.press/v3/brown08a.html},
abstract = {The first Causality Challenge competition posted several causal discovery problems that require researchers to employ the full arsenal of state-of-the-art causal discovery methods, while prompting the development of new ones. Our approach used the formalism of Causal Bayesian Networks to model and induce causal relations and to make predictions about the effects of the manipulation of the variables. Using state-of-the-art, under development, or newly invented methods specifically for the purposes of the competition, we addressed the following problems in turn in order to build and evaluate a model: (a) finding the Markov Blanket of the target even under some non-faithfulness conditions (e.g., parity functions), (b) reducing the problems to a size manageable by subsequent algorithms, (c) identifying and orienting the network edges, (d) identifying causal edges (i.e., not confounded), and (e) selecting the causal Markov Blanket of the target in the manipulated distribution. The results of the competition illustrate some of the strengths and weaknesses of the state-of-the-art of causal discovery methods and point to new directions in the field. An implementation of our approach is available at http://www.dsl-lab.org for use by other researchers.}
}
@InProceedings{chang08a,
title = {Feature Ranking Using Linear SVM},
author = {Yin-Wen Chang and Chih-Jen Lin},
booktitle = {Proceedings of the Workshop on the Causation and Prediction Challenge at WCCI 2008},
pages = {53--64},
year = {2008},
editor = {Isabelle Guyon and Constantin Aliferis and Greg Cooper and André Elisseeff and Jean-Philippe Pellet and Peter Spirtes and Alexander Statnikov},
volume = {3},
series = {Proceedings of Machine Learning Research},
address = {Hong Kong},
month = {03--04 Jun},
publisher = {PMLR},
pdf = {http://proceedings.mlr.press/v3/chang08a/chang08a.pdf},
url = {http://proceedings.mlr.press/v3/chang08a.html},
abstract = {Feature ranking is useful to gain knowledge of data and identify relevant features. This article explores the performance of combining linear support vector machines with various feature ranking methods, and reports the experiments conducted when participating the Causality Challenge. Experiments show that a feature ranking using weights from linear SVM models yields good performances, even when the training and testing data are not identically distributed. Checking the difference of Area Under Curve (AUC) with and without removing each feature also gives similar rankings. Our study indicates that linear SVMs with simple feature rankings are effective on data sets in the Causality Challenge.}
}
@InProceedings{nikulin08a,
title = {Random Sets Approach and its Applications},
author = {Vladimir Nikulin},
booktitle = {Proceedings of the Workshop on the Causation and Prediction Challenge at WCCI 2008},
pages = {65--76},
year = {2008},
editor = {Isabelle Guyon and Constantin Aliferis and Greg Cooper and André Elisseeff and Jean-Philippe Pellet and Peter Spirtes and Alexander Statnikov},
volume = {3},
series = {Proceedings of Machine Learning Research},
address = {Hong Kong},
month = {03--04 Jun},
publisher = {PMLR},
pdf = {http://proceedings.mlr.press/v3/nikulin08a/nikulin08a.pdf},
url = {http://proceedings.mlr.press/v3/nikulin08a.html},
abstract = {The random sets approach is heuristic in nature and has been inspired by the growing speed of computations. For example, we can consider a large number of classifiers where any single classifier is based on a relatively small subset of randomly selected features or random sets of features. Using cross-validation we can rank all random sets according to the selected criterion, and use this ranking for further feature selection. Another application of random sets was motivated by the huge imbalanced data, which represent significant problem because the corresponding classifier has a tendency to ignore patterns with smaller representation in the training set. Again, we propose to consider a large number of balanced training subsets where representatives from both patterns are selected randomly. The above models demonstrated competitive results in two data mining competitions.}
}
@InProceedings{saeed08a,
title = {Bernoulli Mixture Models for Markov Blanket Filtering and Classification},
author = {Mehreen Saeed},
booktitle = {Proceedings of the Workshop on the Causation and Prediction Challenge at WCCI 2008},
pages = {77--91},
year = {2008},
editor = {Isabelle Guyon and Constantin Aliferis and Greg Cooper and André Elisseeff and Jean-Philippe Pellet and Peter Spirtes and Alexander Statnikov},
volume = {3},
series = {Proceedings of Machine Learning Research},
address = {Hong Kong},
month = {03--04 Jun},
publisher = {PMLR},
pdf = {http://proceedings.mlr.press/v3/saeed08a/saeed08a.pdf},
url = {http://proceedings.mlr.press/v3/saeed08a.html},
abstract = {This paper presents the use of Bernoulli mixture models for Markov blanket filtering and classification of binary data. Bernoulli mixture models can be seen as a tool for partitioning an n-dimensional hypercube, identifying regions of high data density on the corners of the hypercube. Once Bernoulli mixture models are computed from a training dataset we use them for determining the Markov blanket of the target variable. An algorithm for Markov blanket filtering was proposed by Koller and Sahami (1996), which is a greedy search method for feature subset selection and it outputs an approximation to the optimal feature selection criterion. However, they use the entire training instances for computing the conditioning sets and have to limit the size of these sets for computational efficiency and avoiding data fragmentation. We have adapted their algorithm to use Bernoulli mixture models instead, hence, overcoming the short comings of their algorithm and increasing the efficiency of this algorithm considerably. Once a feature subset is identified we perform classification using these mixture models. We have applied this algorithm to the causality challenge datasets. Our prediction scores were ranked fourth on SIDO and our feature scores were ranked the best for test sets 1 and 2 of the same dataset.}
}
@InProceedings{yin08a,
title = {Partial orientation and local structural learning of causal networks for prediction},
author = {Jianxin Yin and You Zhou and Changzhang Wang and Ping He and Cheng Zheng and Zhi Geng},
booktitle = {Proceedings of the Workshop on the Causation and Prediction Challenge at WCCI 2008},
pages = {93--105},
year = {2008},
editor = {Isabelle Guyon and Constantin Aliferis and Greg Cooper and André Elisseeff and Jean-Philippe Pellet and Peter Spirtes and Alexander Statnikov},
volume = {3},
series = {Proceedings of Machine Learning Research},
address = {Hong Kong},
month = {03--04 Jun},
publisher = {PMLR},
pdf = {http://proceedings.mlr.press/v3/yin08a/yin08a.pdf},
url = {http://proceedings.mlr.press/v3/yin08a.html},
abstract = {For a prediction problem of a given target feature in a large causal network under external interventions, we propose in this paper two partial orientation and local structural learning (POLSL) approaches, Local-Graph and PCD-by-PCD (where PCD denotes Parents, Children and some Descendants). The POLSL approaches are used to discover the local structure of the target and to orient edges connected to the target without discovering a global causal network. Thus they can greatly reduce computational complexity of structural learning and improve power of statistical tests. This approach is stimulated by the challenge problems proposed in IEEE World Congress on Computational Intelligence (WCCI2008) competition workshop. For the cases with and without external interventions, we select different feature sets to build prediction models. We apply the L1 penalized logistic regression model to the prediction. For the case with noise and calibrant features in microarray data, we propose a two-stage filter to correct global and local patterns of noise.}
}
@InProceedings{cawley09a,
title = {Causal & Non-Causal Feature Selection for Ridge Regression},
author = {Gavin C. Cawley},
booktitle = {Proceedings of the Workshop on the Causation and Prediction Challenge at WCCI 2008},
pages = {107--128},
year = {2008},
editor = {Isabelle Guyon and Constantin Aliferis and Greg Cooper and André Elisseeff and Jean-Philippe Pellet and Peter Spirtes and Alexander Statnikov},
volume = {3},
series = {Proceedings of Machine Learning Research},
address = {Hong Kong},
month = {03--04 Jun},
publisher = {PMLR},
pdf = {http://proceedings.mlr.press/v3/cawley09a/cawley09a.pdf},
url = {http://proceedings.mlr.press/v3/cawley09a.html},
abstract = {In this paper we investigate the use of causal and non-causal feature selection methods for linear classifiers in situations where the causal relationships between the input and response variables may differ between the training and operational data. The causal feature selection methods investigated include inference of the Markov Blanket and inference of direct causes and of direct effects. The non-causal feature selection method is based on logistic regression with Bayesian regularisation using a Laplace prior. A simple ridge regression model is used as the base classifier, where the ridge parameter is efficiently tuned so as to minimise the leave-one-out error, via eigen-decomposition of the data covariance matrix. For tasks with more features than patterns, linear kernel ridge regression is used for computational efficiency. Results are presented for all of the WCCI-2008 Causation and Prediction Challenge datasets, demonstrating that, somewhat surprisingly, causal feature selection procedures do not provide significant benefits in terms of predictive accuracy over non-causal feature selection and/or classification using the entire feature set.}
}