A Conditional Dependence Adjusted Weights of Evidence Model

One of the key assumptions in weights of evidence (WE) modelling is that the predictor patterns have to be conditionally independent. When this assumption is violated, WE posterior probability estimates are likely to be biased upwards. In this paper, a formal expression for the bias of the contrasts will be derived. It will be shown that this bias has an intuitive and convenient interpretation. A modified WE model will then be developed, where the bias is corrected using the correlation structure of the predictor patterns. The new model is termed the conditional dependence adjusted weights of evidence (CDAWE) model. It will be demonstrated via a simulation study that the CDAWE model significantly outperforms the existing WE model when conditional independence is violated, and it is on par with logistic regression, which does not assume conditional independence. Furthermore, it will be argued that, in the presence of conditional dependence between predictor patterns, weights variance estimates from WE are likely to understate the true level of uncertainty. It will be argued that weights variance estimates from CDAWE, which are also bias-corrected, can properly address this issue.

[1]  B. Efron.  Regression and ANOVA with Zero-One Data: Measures of Residual Variation , 1978 .

[2]  Jan Nyssen,et al.  Spatial patterns of old, deep-seated landslides: a case-study in the northern Ethiopian highlands , 2009 .

[3]  John DiNardo,et al.  Econometric methods. 4th ed. , 1997 .

[4]  John C. Davis,et al.  Computers in geology---25 years of progress , 1993 .

[5]  T. Amemiya.  QUALITATIVE RESPONSE MODELS: A SURVEY , 1981 .

[6]  Q. Cheng,et al.  Conditional Independence Test for Weights-of-Evidence Modeling , 2002 .

[7]  E. Ziegel.  Introduction to the Theory and Practice of Econometrics , 1989 .

[8]  D J Vining,et al.  Receiver operating characteristic curves: a basic understanding. , 1992, Radiographics : a review publication of the Radiological Society of North America, Inc.

[9]  DATA INTEGRATION USING WEIGHTS OF EVIDENCE MODEL: APPLICATIONS IN MAPPING MINERAL RESOURCE POTENTIALS , 2002 .

[10]  F. Agterberg,et al.  Weights of evidence modelling: a new approach to mapping mineral potential , 1990 .

[11]  Graeme F. Bonham-Carter,et al.  Measuring the Performance of Mineral-Potential Maps , 2005 .

[12]  F. Agterberg,et al.  Statistical Pattern Integration for Mineral Exploration , 1990 .

[13]  M. Zweig,et al.  Receiver-operating characteristic (ROC) plots: a fundamental evaluation tool in clinical medicine. , 1993, Clinical chemistry.

[14]  G. Rawat,et al.  Weights of evidence modelling for landslide hazard zonation mapping in part of Bhagirathi valley, Uttarakhand , 2007 .

[15]  Guocheng Pan,et al.  A Comparative Analysis of Favorability Mappings by Weights of Evidence, Probabilistic Neural Networks, Discriminant Analysis, and Logistic Regression , 2003 .

[16]  Ru-Hua Song,et al.  Modeling the potential distribution of shallow-seated landslides using the weights of evidence method and a logistic regression model： a case study of the Sabae Area, Japan , 2008 .