Our goal is to explore the feasibility and usefulness of using a combination of covering arrays and machine learning models for predicting results of an agent- based simulation model within the vast parameter value combination space. The challenge is to select parameter values that are representative of the overall behavior of the model, so that we can train the machine learning model to be able to correctly predict behavior on previously untested areas of the parameter space. We have chosen Wilensky's Heat Bugs model in NetLogo for our study. It is a simple model, amenable to quick data generation, with a limited number of outputs to predict, and with emergent behavior. This model therefore allows exploration of this new approach.We utilize covering arrays to reduce the parameter value space systematically, run the model for each parameter set in the 2-way and 3-way covering arrays, train a random forest model on the 2-way data (33, 351 parameter combinations), and test its ability to predict the outcome of the simulation on the significantly larger 3-way data that was not seen during the training of the model (3, 971, 955 parameter combinations).
About this Dataset
| Title | Predicting ABM Results with Covering Arrays and Random Forests |
|---|---|
| Description | Our goal is to explore the feasibility and usefulness of using a combination of covering arrays and machine learning models for predicting results of an agent- based simulation model within the vast parameter value combination space. The challenge is to select parameter values that are representative of the overall behavior of the model, so that we can train the machine learning model to be able to correctly predict behavior on previously untested areas of the parameter space. We have chosen Wilensky's Heat Bugs model in NetLogo for our study. It is a simple model, amenable to quick data generation, with a limited number of outputs to predict, and with emergent behavior. This model therefore allows exploration of this new approach.We utilize covering arrays to reduce the parameter value space systematically, run the model for each parameter set in the 2-way and 3-way covering arrays, train a random forest model on the 2-way data (33, 351 parameter combinations), and test its ability to predict the outcome of the simulation on the significantly larger 3-way data that was not seen during the training of the model (3, 971, 955 parameter combinations). |
| Modified | 2023-04-20 00:00:00 |
| Publisher Name | National Institute of Standards and Technology |
| Contact | mailto:[email protected] |
| Keywords | agent-based modeling · machine learning · calibration |
{
"identifier": "ark:\/88434\/mds2-3002",
"accessLevel": "public",
"contactPoint": {
"hasEmail": "mailto:[email protected]",
"fn": "M S Raunak"
},
"programCode": [
"006:045"
],
"landingPage": "https:\/\/data.nist.gov\/od\/id\/mds2-3002",
"title": "Predicting ABM Results with Covering Arrays and Random Forests",
"description": "Our goal is to explore the feasibility and usefulness of using a combination of covering arrays and machine learning models for predicting results of an agent- based simulation model within the vast parameter value combination space. The challenge is to select parameter values that are representative of the overall behavior of the model, so that we can train the machine learning model to be able to correctly predict behavior on previously untested areas of the parameter space. We have chosen Wilensky's Heat Bugs model in NetLogo for our study. It is a simple model, amenable to quick data generation, with a limited number of outputs to predict, and with emergent behavior. This model therefore allows exploration of this new approach.We utilize covering arrays to reduce the parameter value space systematically, run the model for each parameter set in the 2-way and 3-way covering arrays, train a random forest model on the 2-way data (33, 351 parameter combinations), and test its ability to predict the outcome of the simulation on the significantly larger 3-way data that was not seen during the training of the model (3, 971, 955 parameter combinations).",
"language": [
"en"
],
"distribution": [
{
"downloadURL": "https:\/\/data.nist.gov\/od\/ds\/mds2-3002\/2way-toshare.csv",
"format": "Predicting ABM Results with Covering Arrays and Random Forests",
"description": "Predicting ABM Results with Covering Arraysand Random Forests",
"mediaType": "text\/csv",
"title": "Predicting ABM Results with Covering Arrays and Random Forests"
}
],
"bureauCode": [
"006:55"
],
"modified": "2023-04-20 00:00:00",
"publisher": {
"@type": "org:Organization",
"name": "National Institute of Standards and Technology"
},
"theme": [
"Information Technology:Data and informatics"
],
"keyword": [
"agent-based modeling \u00b7 machine learning \u00b7 calibration"
]
}
