Friday, March 29, 2019

Drafting Agent-Based Modeling into Basketball Analytics

http://scs.org/springsim/
Readers of this blog might find this post a little  out of left field (sorry I could not find a better analogy) as it about basketball and therefore the ball is in your court if you want to keep reading.

At the upcoming SpringSim conference Matthew Oldham and myself  just had a paper accepted entitled "Drafting Agent-Based Modeling into Basketball Analytics" where we take a shot at modeling basketball. Why you might ask? The rational is that sports analytics (SA) is a multi-million dollar industry but to date little attention has been given to agent-based modeling (ABM) even though sports can be viewed as a complex adaptive system (Matthew on his site has a great write up of this). To explore this notion we built an agent-based model (utilizing NetLogo 3D) which captures the basic dynamics of a basketball game. In order to calibrate the processes within the model we utilized 17 seasons (2000 to 2016) of individual game data from the National Basketball Association (NBA). The data collected included; game scores, winning margins, field goal attempts, the percentage of field goals made, rebounds, steals, and turnovers. From the NBA game data, density functions were calculated to aid calibrating certain aspects of the model.  Via a set of experiments, the model indicates that an increased belief in the franchise player (think Michael Jordan) leads to increased scoring action, but a belief in the hot-hand had a minor effect. This results comes from the ability of agent-based models to identify the micro-interaction of agents responsible for generating system level outcomes and thereby, demonstrating the utility of ABM to SA.

Below, you a can read the abstract of the paper, along with some figures outlining the play cycle in the model, some some results of the varying NBA game metrics compared to the model, a movie of the graphical user interface of the model during a representative game. Finally at the end of the post you can find link to the model and the actual paper. 
The growth of sports analytics (SA) has raised numerous research topics across a variety of sports, including basketball. Agent-based modeling (ABM) has great potential to assist and inform SA, but to date it has not been utilized. To support the use of ABM in SA, a model of a basketball game, which considers most fundamentals of play, is presented. Additionally, player behavior is partially predicated on assessing the length of a player’s shooting streak (testing the “hot-hand” effect) and the consideration a team gives to a streak and their franchise player. The model’s output is used to calibrate and validate it against statistics from the National Basketball Association (NBA). Via a set of experiments, the model indicates that an increased belief in the franchise player leads to increased scoring action, but a belief in the hot-hand a minor effect. Thereby, demonstrating the utility of ABM to SA, thus opening a new research field.

Keywords: agent-based modeling, sports analytics, hot-hand effect.
Figure 2: The play cycle of the model.

Figure 3: Distribution of the varying NBA game metrics compared to the model.




The model along with a detailed Overview, Design concepts, and Details (ODD) document can be found at: https://tinyurl.com/ABMBasketBall.

Full Reference:
Oldham, M and Crooks, A.T. (2019) Drafting Agent-Based Modeling into Basketball Analytics, 2019 Spring Simulation Conference (SpringSim’19), Tucson, AZ. (pdf)

Friday, January 18, 2019

Agent-based Modelling and Geographical Information Systems: A Practical Primer

Its been a long time in the making but now "Agent-Based Modelling and Geographical Information Systems: A Practical Primer" has been published by Sage. We (Nicolas Malleson, Ed Manley, Alison Heppenstall and myself) approached this book from two standpoints. First, to provide a synthesis of the underpinning ideas, techniques and frameworks for integrating agent-based modelling and geographical information systems (GIS). Second, building on our experiences of teaching at various levels, to provide a practical set of information for the development of agent-based models for geographical systems.


From these two standpoints we have developed a book that provides a practical primer in the integration of agent-based modelling and geographical information systems. In outlining the subject we cover many examples of geographical phenomena, from linking the individual movements of pedestrians to aggregate patterns of urban growth, to the integration of social networks into modelling mobility. Through this text, we hope  the reader will understand how the field has developed, how agent-based models are different from other modelling approaches, and the future challenges we see lying ahead.
By using sample code and data (all of which can be found on the accompanying website https://www.abmgis.org/) we provide the reader with many of the basic building blocks for constructing agent-based models linked to geographical information systems. Throughout the book we use the software package NetLogo, as it provides an easy route to learn and build agent-based models (although in the appendix we provide links to other models created in other platforms).

For more information visit https://www.abmgis.org/ and if you wish to buy a copy you can: Amazon or Sage Publishing. We hope you enjoy it. 
Figure 6.1 Abstracting from the real world to a series of layers to be used in the artificial world
upon which the agent-based model is based.
Full Reference:
Crooks, A.T., Malleson, N., Manley, E. and Heppenstall, A.J. (2019), Agent-based Modelling and Geographical Information Systems: A Practical Primer, Sage, London, UK.

Wednesday, January 02, 2019

Models from Teaching CSS Fall 2018

Most of the time when I teach a class instead of setting a final exam, I ask the students to carryout an end of semester research project. In my Introduction to Computational Social Science class this project entails the development of a computational model in an area of  interest to the student . The aim of this exercise is to cement what the students have (hopefully) learnt during the semester. I.e.: 
  • to understand the motivation for the use of computational models in social science theory and research; 
  • to learn about the variety of CSS research programs across the social science disciplines; 
  • to understand the distinct contribution that CSS can make by providing specific insights about society, social phenomena at multiple scales, and the nature of social complexity.
Below you can see some of the outputs from these projects this last fall. The models range in type from agent-based models, microsimulation to system dynamics models applied to a variety of topics from voting and political parties, the peer effects of students, urban decline, employment growth and rise and fall of civilizations and many other topics along the way.


Wednesday, December 05, 2018

Detecting and Mapping Slums using Open Data

Urban and slum areas in Nairobi (False composite image created
by stacking image bands 7, 6 and 4 from the Landsat 8 satellite.
Turning back to slums, we just published paper entitled "Detecting and Mapping Slums using Open Data: A Case Study in Kenya" in the International Journal of Digital Earth. This work builds and extends our previous research on using new sources of data to explore the slum settlements in 3 cities in Kenya (i.e. Nairobi, Mombasa and Kisumu). Specifically, we examine how the fusion of Volunteered Geographical Information, Social Media, and other open data sources can complement remote sensing imagery in supporting slum detection, mapping and monitoring. 

We do this by using data mining tools (e.g. logistic regression, discriminant analysis and the See5 decision tree), to develop context-sensitive definitions for slums based on location, as well as for testing the generalizability of indicators and derived slum models. The end result is an indicator database for slums using open sources of physical and socio-economic data that can be used to characterize slum settlements. If you wish to know more, below we provide the abstract to the paper along with some of the figures and the full citation with a link to the paper itself.

Abstract:
The worldwide slum population currently stands at over one billion, with substantial growth expected in the coming decades. Traditionally, slums have been mapped using information derived mainly from either physical indicators using remote sensing data, or socio-economic indicators using census data. Each data source on its own provides only a partial view of slums, an issue further compounded by data poverty in less developed countries. To overcome such issues, this paper explores the fusion of traditional with emerging open data sources and data mining tools to identify additional indicators that can be used to detect and map the presence of slums, map their footprint, and map their evolution. Towards this goal, we develop an indicator database for slums using open sources of physical and socio-economic data that can be used to characterize slum settlements. Using this database, we then leverage data mining techniques to identify the most suitable combination of these indicators for mapping slums. Using three cities in Kenya as test cases, results show that the fusion of these data can improve the mapping accuracy of slums. These results suggest that the proposed approach can provide a viable solution to the emerging challenge of monitoring the growth of slums.
Keywords: Slums; Remote Sensing; Socio-economic; Urban sustainability; Data mining; Kenya

Study areas in Kenya

Methodology workflow

Distribution of positive classified cases for slums for (a) logistic regression, (b) discriminant analysis and (c) the See5 decision tree.
Full Reference:
Mahabir, R., Agouris, P., Stefanidis, A., Croitoru, A. and Crooks, A.T. (2018), Detecting and Mapping Slums using Open Data: A Case Study in Kenya, International Journal of Digital Earth. DOI: https://doi.org/10.1080/17538947.2018.1554010. (pdf)

Bots in Social Networks

Recently our research has started to dig deeper into social media, especially how bots diffuse information in online social networks (OSNs). To this end at the  7th International Conference on Complex Networks and Their Applications we had a paper entitled: Bots in Nets: Empirical Comparative Analysis of Bot Evidence in Social Networks

In this paper we present a framework to characterize the pervasiveness and relative importance of bots in various OSNs conversations of three significant global events in 2016. In total, we harvested more than 30 million tweets from the U.S. Presidential Election, the Ukrainian Conflict and Turkish Political Censorship and compared the conversational patterns of bots and humans within each event. The results from this analysis showed that although Twitter participants identified as social bots comprised only 0.28% of all OSN users in this study, they accounted for a significantly large portion of prominent centrality rankings across the three conversations. If you want to know more about this new work, below we provide the abstract to the paper, a selection of figures and tables (including our methodology, some summary information about our data corpus and some of the results). Finally at the bottom of this post we have the full reference and a link to the paper.

Abstract:
The emergence of social bots within online social networks (OSNs) to diffuse information at scale has given rise to many efforts to detect them. While methodologies employed to detect the evolving sophistication of bots continue to improve, much work can be done to characterize the impact of bots on communication networks. In this study, we present a framework to describe the pervasiveness and relative importance of participants recognized as bots in various OSN conversations. Specifically, we harvested over 30 million tweets from three major global events in 2016 (the U.S. Presidential Election, the Ukrainian Conflict and Turkish Political Censorship) and compared the conversational patterns of bots and humans within each event. We further examined the social network structure of each conversation to determine if bots exhibited any particular network influence, while also determining bot participation in key emergent network communities. The results showed that although participants recognized as social bots comprised only 0.28% of all OSN users in this study, they accounted for a significantly large portion of prominent centrality rankings across the three conversations. This includes the identification of individual bots as top-10 influencer nodes out of a total corpus consisting of more than 2.8 million nodes. 

Keywords: bots, online social networks, social network analysis.
Fig. 1. Overall methodology to analyze bot evidence across multiple Twitter OSN conversations.
Table 1. Harvested Twitter Corpus Overview
Fig. 5. Correlation of centrality measures for select centrality comparisons: (a) U.S. Election eigenvector versus betweenness analysis, (b) Ukraine Conflict eigenvector versus betweenness analysis and (c)Ukraine Conflict eigenvector versus degree analysis.
Table 2. Bot density of largest emergent communities.

Full Reference:
Schuchard, R., Crooks, A.T., Stefanidis, A.  and Croitoru, A. (2018), Bots in Nets: Empirical Comparative Analysis of Bot Evidence in Social Networks, in Aiello, L.M., Cherifi, C. Cherifi, H., Lambiotte, R., LiĆ³, P. and Rocha, L.M. (eds.), Volume 2, Proceedings of the 7th International Conference on Complex Networks and Their Applications, Cambridge, United Kingdom, Springer, pp 424-436. (pdf)