Register now After registration you will be able to apply for this opportunity online.
This opportunity is not published. No applications will be accepted.
Machine Learning for Short-Term Crime Prediction
The aim of this master thesis is to develop a crime prediction model on street level for a Swiss city.
Keywords: crime prediction, machine learning, data mining, criminology, forecasting, big data, learning to rank, predictive analytics, urban computing, SQL, python, R
The research methodology is expected to comprise:
- Literature review in the relevant data mining and quantitative criminology domains
- Identification of relevant data sources (census data, street network and other urban structures, OpenStreetMap, Foursquare, forecast.io, etc.), collection and consolidation of the data on street level and different temporal granularities
- Implementation and evaluation of a machine learning model, and comparison to an established baseline model. (e.g. identifying top streets at risk for burglary in the next time period using some learning-to-rank methods and comparing them to current network-based prospective crime mapping).
- Optional: prototype development of a web-based presentation of the results.
Teasers:
- current predictive policing software packages used operationally in USA or Europe include PredPol and HunchLab, watch this video: https://www.youtube.com/watch?v=YxvyeaL7NEM
- relevant criminological theories and data sources for crime prediction: https://vimeo.com/145402491
You should meet following requirements:
- Sound knowledge of SQL. Experience working with spatial data is a plus (e.g. PostGIS).
- Programming/scripting experience (Python or R preferred).
- Data Mining, applied Machine Learning, or Econometrics skills are mandatory (e.g. scikit-learn in python or glmnet in R).
- Interest in the crime and computational social science domains.
- Proficiency in English.
- Optional: full-stack development. Frameworks like Flask, Django, D3js for deploying the crime prediction results.
Ideal candidate would have a background background in math/computer science or another quantitative discipline and experience with ML.
The research methodology is expected to comprise: - Literature review in the relevant data mining and quantitative criminology domains - Identification of relevant data sources (census data, street network and other urban structures, OpenStreetMap, Foursquare, forecast.io, etc.), collection and consolidation of the data on street level and different temporal granularities - Implementation and evaluation of a machine learning model, and comparison to an established baseline model. (e.g. identifying top streets at risk for burglary in the next time period using some learning-to-rank methods and comparing them to current network-based prospective crime mapping). - Optional: prototype development of a web-based presentation of the results.
Teasers: - current predictive policing software packages used operationally in USA or Europe include PredPol and HunchLab, watch this video: https://www.youtube.com/watch?v=YxvyeaL7NEM - relevant criminological theories and data sources for crime prediction: https://vimeo.com/145402491
You should meet following requirements: - Sound knowledge of SQL. Experience working with spatial data is a plus (e.g. PostGIS). - Programming/scripting experience (Python or R preferred). - Data Mining, applied Machine Learning, or Econometrics skills are mandatory (e.g. scikit-learn in python or glmnet in R). - Interest in the crime and computational social science domains. - Proficiency in English. - Optional: full-stack development. Frameworks like Flask, Django, D3js for deploying the crime prediction results. Ideal candidate would have a background background in math/computer science or another quantitative discipline and experience with ML.
The goal of the thesis is to explore the frontiers of a crime prediction model on street level in terms of exploited data sources and employed machine learning techniques. The research group has already experience with modeling crime on US and Swiss data at different granularity levels. You have the chance to do highly relevant research and become a co-author of a scientific paper in a top conference or journal.
The goal of the thesis is to explore the frontiers of a crime prediction model on street level in terms of exploited data sources and employed machine learning techniques. The research group has already experience with modeling crime on US and Swiss data at different granularity levels. You have the chance to do highly relevant research and become a co-author of a scientific paper in a top conference or journal.
Please send your application with [Master Thesis – crime prediction ML] in the subject line to: Cristina Kadar, PhD Candidate Chair of Information Management (Professor Elgar Fleisch) D-MTEC, ETH Zurich Homepage: http://www.im.ethz.ch/people/ckadar.html Email: ckadar {at} ethz {dot} ch
Include CV and transcript of records, together with your motivation to work on the topic.
Please send your application with [Master Thesis – crime prediction ML] in the subject line to: Cristina Kadar, PhD Candidate Chair of Information Management (Professor Elgar Fleisch) D-MTEC, ETH Zurich Homepage: http://www.im.ethz.ch/people/ckadar.html Email: ckadar {at} ethz {dot} ch Include CV and transcript of records, together with your motivation to work on the topic.