Deep Neural Network-Based Crime Prediction Using Twitter Data

Deep Neural Network-Based Crime Prediction Using Twitter Data

Chamith Sandagiri, Banage T. G. S. Kumara, Banujan Kuhaneswaran
Copyright: © 2021 |Pages: 16
DOI: 10.4018/IJSSOE.2021010102
(Individual Articles)
No Current Special Offers


Crimes have affected the quality of life and economic growth of the country badly. The authors can identify the crime patterns and predict the crimes by detecting and analyzing the historical data. However, some crimes are unregistered and unsolved due to a lack of evidence. Thus, detecting crimes is a still challenging task. Individuals can use social media like Twitter to detect crime-related activities. Because Twitter users sometimes convey messages related to their surrounding environment, this paper proposed a machine learning approach to predict crimes. The proposed framework consists of three modules: data (tweet) collecting, detecting crimes, and predicting crime. Long short-term memory (LSTM) neural network model was used as a proposed approach for crime prediction. Experimental results found that by achieving the highest precision of 82.5%, precision of 86.4%, and recall of 80.4%, the proposed LSTM-based approach worked better than the other approaches.
Article Preview


The main goal of every country is national security issues. Criminology experiments rely on criminal traits recognition. The use of data mining methods may help to define this. Crime analysis is a criminal law enforcement function that includes the analysis of patterns and trends in crime and disorder. Crime analysis is part of criminology. In today's world, criminals are technologically advanced and sometimes share their feelings on the internet. The phenomenal growth of the World Wide Web has led more people to share their views online. (Bolla, 2014). Advances in crime detection are important to optimize the distribution of finite resources. However, traditional crime modeling methods have some disadvantaged in capturing crime events, as the theoretical model does not include criminal forecasting variables. Many causes can influence potential reports of crime rather than hot spots. However, a more comprehensive forensic investigation is required to select the potential causal factors for crime incidents (X. Chen, Cho, & Jang, 2015). Different forms of crime have different spatial possibilities that need separate analysis. Likewise, the population at risk ranges at varying crime rates and should be taken into account in the same way (Malleson & Andresen, 2015).

The content of social media has detailed qualitative knowledge on the everyday lives of its members based on shared textual evidence. Each text post is compiled to the service provider portal. These threads or posts build an unstructured data format (X. Chen et al., 2015). Twitter and Facebook are also used for reporting crimes and responding to demands for witnesses and facts. There is proof of concept that handheld mass media are used for sensing in urban areas. Many sites use this principle of crowdsourcing to help the police monitor offenders' actions and outsource them to a group of staff to collect input from the public. (Bendler, Brandt, Wagner, & Neumann, 2014). Thus, social media data can be viewed as a good instrument for forecasting crime. We chose Twitter to provide data to improve the advanced crime prediction model (X. Chen et al., 2015). Researchers and application developers get a huge number of Tweets for different data analysis via Streaming API (Abbass, Ali, Ali, Akbar, & Saleem, 2020)

In 2012, it was the first time that social media and crime were put together to make a prediction. This approach to the combination of social media and crime data is built through the topic extraction of themes and linkages with crime occurrence (X. Wang, Gerber, & Brown, 2012). Automatic semantic analysis and Natural Language Processing (NLP) of Twitter data, dimensionality reduction through Latent Dirichlet Allocation (LDA) (Blei, Ng, & Jordan, 2003), and prediction with linear modeling for hit-and-run crimes in Charlottesville, Virginia represented the earliest research on this topic. The second approach points out identifying vehicle descriptions in twitter content to predict crime (Featherstone, 2013). It is a new research area used in social media content with a different method.

Comprehensive research, human activity pattern, and crime analysis & forecast is accessible in all fields. Most of the study also incorporates the subjective interpretation of social media or the real detection of hot spots for crime. We seek to investigate the association between illegal interaction in an urban setting and place and the similarity between various forms of crime. This study seeks to examine what we can infer about social media interactions that forecast the incidence of crimes.

However, it is very difficult to use Twitter as a source of information for crime prediction. Tweets are notorious for (un)intentional misprints, on-the-fly word invention, using symbols, and syntactic structures which often challenge even the most straightforward computational treatment (e.g. word boundary identification) (Gerber, 2014).

The remainder of this work is structured as follows. In the second section, we provide an overview of recent work in related research areas. From the literature review, we can derive the research gap that is attempted by this work. We provide the preliminaries which include the motivation for the research in the third section. The applicability of the LSTM model to assess the explanatory power of Social Media is judged in the fourth section. The fifth section states the findings of our research. This work closes with a conclusion in the sixth section and gives an outlook on future research.

Complete Article List

Search this Journal:
Volume 13: 1 Issue (2024): Forthcoming, Available for Pre-Order
Volume 12: 2 Issues (2022): 1 Released, 1 Forthcoming
Volume 11: 2 Issues (2021)
Volume 10: 2 Issues (2020)
Volume 9: 2 Issues (2019)
Volume 8: 4 Issues (2018)
Volume 7: 4 Issues (2017)
Volume 6: 4 Issues (2016)
Volume 5: 4 Issues (2015)
Volume 4: 4 Issues (2014)
Volume 3: 4 Issues (2012)
Volume 2: 4 Issues (2011)
Volume 1: 4 Issues (2010)
View Complete Journal Contents Listing