Incorporating model identifiability into equation discovery of ODE systems

Equation discovery is a machine learning technique that tries to automate the discovery of equations from measured data. In this contribution an equation discovery system based on genetic programming was developed in order to generate mechanistic models for systems described by ordinary differential equations. A problem often encountered with automatic model generation is that overly complex models are generated that "overfit" the measured data. This issue was addressed by incorporating a model identifiability measure (expressing which fraction of the model parameters can be given a unique value given the available data) into the fitness function of the individuals. Using noisy artificially generated data for a river water quality example case, it was shown that the developed system was able to generate model equations that fitted the data well and were also fully identifiable. Correct model equations were generated when starting from a model with minimum prior knowledge but also when starting from an overly complex model. As such, it was demonstrated that the developed equation discovery system is able to generate models with optimal complexity with regard to the available data.

[1]  A. Wood,et al.  Towards the systematic simplification of mechanistic models , 2006 .

[2]  Saso Dzeroski,et al.  Inductive process modeling , 2008, Machine Learning.

[3]  Sašo Džeroski,et al.  Constructing a library of domain knowledge for automated modelling of aquatic ecosystems , 2006 .

[4]  D J W De Pauw,et al.  Identifiability analysis and improvement of a tree water flow and storage model. , 2008, Mathematical biosciences.

[5]  D. Savić,et al.  A symbolic data-driven technique based on evolutionary polynomial regression , 2006 .

[6]  Pat Langley,et al.  Using equation discovery to revise an Earth ecosystem model of the carbon net production , 2003 .

[7]  Pat Langley,et al.  Reducing overfitting in process model induction , 2005, ICML '05.

[8]  David J. Murray-Smith,et al.  Nonlinear model structure identification using genetic programming , 1998 .

[9]  Kazumi Saito,et al.  Robust Induction of Process Models from Time-Series Data , 2003, ICML.

[10]  Kazumi Saito,et al.  Computational Revision of Quantitative Scientific Models , 2001, Discovery Science.

[11]  Earle B. Phelps,et al.  A Study of the Pollution and Natural Purification of the Ohio River , 1958 .

[12]  Pat Langley,et al.  Constructing explanatory process models from biological data and knowledge , 2006, Artif. Intell. Medicine.

[13]  Saso Dzeroski,et al.  Integrating Knowledge-Driven and Data-Driven Approaches to Modeling , 2006, EnviroInfo.

[14]  Grigorios N. Beligiannis,et al.  Nonlinear model structure identification of complex biomedical data using a genetic-programming-based technique , 2005, IEEE Transactions on Instrumentation and Measurement.

[15]  Sean Luke,et al.  A Comparison of Bloat Control Methods for Genetic Programming , 2006, Evolutionary Computation.

[16]  John A. Nelder,et al.  A Simplex Method for Function Minimization , 1965, Comput. J..

[17]  J. Nash,et al.  River flow forecasting through conceptual models part I — A discussion of principles☆ , 1970 .

[18]  Martin Stacey,et al.  Scientific Discovery: Computational Explorations of the Creative Processes , 1988 .

[19]  Peter Reichert,et al.  Practical identifiability of ASM2d parameters--systematic selection and tuning of parameter subsets. , 2002, Water research.

[20]  Guido Buzzi-Ferraris,et al.  A new sequential experimental design procedure for discriminating among rival models , 1983 .

[21]  Sašo Džeroski,et al.  Modelling and prediction of phytoplankton growth with equation discovery , 1998 .