Publicação
Time series forecasting on crime data in Amsterdam for a software company
| dc.contributor.advisor | Mendes, Jorge Morais | |
| dc.contributor.advisor | Hoekstra, Vincent | |
| dc.contributor.author | Singh, Prakash | |
| dc.date.accessioned | 2019-01-18T17:33:59Z | |
| dc.date.available | 2019-01-18T17:33:59Z | |
| dc.date.issued | 2018-12-21 | |
| dc.description | Internship Report presented as the partial requirement for obtaining a Master's degree in Data Science and Advanced Analytics | pt_PT |
| dc.description.abstract | In recent years, there have been many discussions of data mining technology implementation in the fight against terrorism and crime. Sentient as a software company has been supporting the police for years by applying data mining techniques in the DataDetective application (Sentient, 2017). Experimenting with various types of predictive model solutions, selecting the most efficient and promising solution are the objectives of this internship. Initially, extended literatures were reviewed in the field of data mining, crime analysis and crime data mining. Sentient provided 7 years of crime data which was aggregated on daily basis to create a univariate dataset. Also, an incidence type daily aggregation was done to create a multivariate dataset. The prediction length for each solution was 7 days. The experiments were divided into two major categories: Statistical models and neural network models. Neural networks outperformed statistical models for the crime data. This paper provides the overview of statistical models and neural network models used. A comparative study of all the models on similar dataset gives a clear picture of their performance on available data and generalization capability. Evidently, the experiments showed that Gated Recurrent units (GRU) produced better prediction in comparison to other models. In conclusion, gated recurrent unit implementation could give benefit to police in predicting crime. Hence, time series analysis using GRU could be a prospective additional feature in DataDetective. | pt_PT |
| dc.identifier.tid | 202150976 | pt_PT |
| dc.identifier.uri | http://hdl.handle.net/10362/57826 | |
| dc.language.iso | eng | pt_PT |
| dc.rights.uri | http://creativecommons.org/licenses/by/4.0/ | pt_PT |
| dc.subject | Time series analysis | pt_PT |
| dc.subject | Forecasting | pt_PT |
| dc.subject | ARIMA | pt_PT |
| dc.subject | Supervised learning | pt_PT |
| dc.subject | Machine learning | pt_PT |
| dc.subject | Predictive models | pt_PT |
| dc.subject | Neural networks | pt_PT |
| dc.subject | Recurrent Neural Networks | pt_PT |
| dc.subject | Convolutional neural networks | pt_PT |
| dc.subject | Relatórios de Estágio | pt_PT |
| dc.title | Time series forecasting on crime data in Amsterdam for a software company | pt_PT |
| dc.type | master thesis | |
| dspace.entity.type | Publication | |
| rcaap.rights | openAccess | pt_PT |
| rcaap.type | masterThesis | pt_PT |
| thesis.degree.name | Mestrado em Métodos Analíticos Avançados | pt_PT |