Optimising Policing: Knowledge Discovery for Targeted Strategies

Version	Summary	Created by	Modification	Content Size	Created at	Operation
1		Marcio Pereira Basilio	--	1873	2025-05-31 05:50:50	\|
2	I've adjusted the description to be more specific.	Marcio Pereira Basilio	+ 67 word(s)	1940	2025-05-31 06:02:21	\| \|
3	format correct	Catherine Yang	+ 3 word(s)	1943	2025-06-03 03:17:33	\|

Version

Summary

Created by

Modification

Content Size

Created at

Operation

Marcio Pereira Basilio

1873

2025-05-31 05:50:50

I've adjusted the description to be more specific.

Marcio Pereira Basilio

+ 67 word(s)

1940

2025-05-31 06:02:21

| |

format correct

Catherine Yang

+ 3 word(s)

1943

2025-06-03 03:17:33

Effective policing is a cornerstone of a well-functioning society, directly impacting public safety and trust. This study's methodology, by integrating text mining (LDA) and multi-criteria decision aid (ELECTRE) techniques, offers a crucial advancement in this domain. By accurately identifying specific local criminal demands from raw emergency call data, police forces can move away from one-size-fits-all approaches, which are often inefficient and resource-intensive. The ability to discern unique crime patterns in different areas—demonstrated by the 40% distinct demands between AISP 5 and AISP 19—underscores the necessity for tailored strategies. For society, this means a more responsive and effective police presence. Resources, often scarce, can be optimally allocated to address the most pressing local criminal issues, leading to tangible reductions in crime rates. This targeted approach, supported by the model's ability to rank policing strategies based on their impact, not only enhances operational efficiency but also fosters a greater sense of security among citizens. Ultimately, by leveraging data to understand and combat crime more intelligently, this research contributes to a safer, more stable, and more trusting relationship between law enforcement and the communities they serve.

1. Introduction

The persistent challenge of crime in modern societies necessitates the continuous evolution of policing strategies. Traditional approaches often fail to address criminal activity's complex and heterogeneous nature, leading to inefficient resource allocation and sub-optimal crime reduction outcomes. This study presents a novel methodological framework to enhance law enforcement agencies' decision-making processes by integrating advanced data analytics with multi-criteria decision-aid techniques. By extracting meaningful insights from large volumes of unstructured emergency call data, the proposed model aims to facilitate the selection and prioritization of policing strategies tailored to the specific criminal demands of diverse geographical areas. This approach moves beyond generic policing models, offering a data-driven path toward more effective and resource-efficient crime prevention and response.

2. Research and Findings

The core objective of this research was to develop a robust method for knowledge discovery within the extensive databases of the 190 emergency service, drawing directly from police incident reports. This method was designed to produce actionable information that precisely identifies local criminal demands, thereby allowing for the ranking of appropriate policing strategies to address these issues. The specific aims were meticulously defined, encompassing the identification of local criminal demands from police incident records, the comprehensive identification of available policing strategies, the discernment of the most frequent demands, the construction of a detailed impact matrix linking policing strategies to criminal demand reduction, and, finally, the systematic ranking of these strategies based on identified criminal demands.

The methodological framework adopted in this study is a sophisticated knowledge discovery approach that seamlessly integrates advanced text mining techniques, specifically Latent Dirichlet Allocation (LDA), with the ELECTRE multi-criteria method. This hybrid methodology offers a powerful lens to analyse and interpret complex criminal data. Data collection was meticulously carried out in close collaboration with the Military Police of the State of Rio de Janeiro, leveraging records of emergency calls processed by the 190 service from 1 January 2013 to 31 December 2016. For empirical analysis, two Integrated Public Safety Areas (AISP) were randomly selected: AISP 5, representing Downtown Rio de Janeiro, and AISP 19, encompassing Copacabana, Rio de Janeiro. The methodological process unfolded through several distinct stages. The initial phase involved the Reading and Transformation of Text, where relevant fields from the extensive database were meticulously extracted and individualized into different text files, forming the corpus for subsequent analysis. This resulted in a substantial dataset comprising 24,286 text files for the 5th Military Police Battalion (BPM) area (AISP 5) and 14,374 files for the 19th BPM area (AISP 19). Following this, the Extraction and Cleaning of Terms stage commenced, where the corpus underwent a rigorous tokenisation process involving the decomposition of text into individual terms. This stage systematically removed symbols, control characters, punctuation, numbers, dates, and diacritics. Furthermore, extensive lists of stopwords (both English and Portuguese), acronyms, meaningless phonemes, military alphabet terms, month designations, police ranks and graduations, and abbreviations of police units were purged to refine the dataset. Critically, specific terms were replaced with standardised synonyms to ensure terminological consistency. A deliberate decision was made to omit the stemming process to preserve crucial semantic meaning, which is vital for accurately differentiating between active and passive agents, determining the progression of an action, and identifying the gender of individuals involved in incidents.

The subsequent stage focused on the Construction of the Document Term Matrix (DTM), which involved categorizing the refined terms and associating them with their respective frequencies of occurrence within the corpus. This process facilitated inferences regarding term proximities and relationships. The corpus from the 19th BPM, initially comprising 14,374 elements, was reduced to 10,427 after preprocessing. Similarly, the 5th BPM corpus, originally with 24,286 elements, was streamlined to 14,786. Both DTMs were ultimately refined to 4,918 elements after removing sparse terms.

The critical phase of Identification of Topics commenced, applying the Latent Dirichlet Allocation (LDA) method with Gibbs Sampling to derive topics pertinent to the 5th and 19th BPM areas. LDA, a generative probabilistic model, assumes that each document within a corpus is composed of a mixture of a limited number of underlying topics, with each topic contributing various associated words. This method effectively identifies clusters of words that frequently co-occur, thereby revealing underlying themes or subjects within subsets of documents. This capability of LDA is pivotal for "knowledge discovery within 190 emergency response databases," transforming unstructured police reports into discernible patterns of criminal activity.

Following topic identification, the Validation of Topics stage was crucial. This involved the development and distribution of questionnaires to 100 experienced police officers specialising in public security. These specialists, predominantly male (85%) with an average age of 40.59 years and an average of 19.35 years of service, validated the coherence of the words within each identified topic in 95% of cases. The process of assigning labels to the topics, based on the most frequently suggested terms by these experts, converted raw data into actionable criminal demands. Concurrently, Identification of Alternatives (Policing Strategies) involved the comprehensive review of both academic literature and the operational practices of the Military Police of the State of Rio de Janeiro, leading to the identification of fourteen distinct policing strategies to be evaluated within the model. Correspondingly, Identification of Criteria (Criminal Demands) involved defining twenty specific criteria, which represent the most prevalent types of crimes, misdemeanours, and urban disorder regularly monitored by the Public Security Institute (ISP).

The empirical foundation of the model was further solidified in the Obtainment of the Alternatives versus Criteria Evaluation Matrix. This matrix, a pivotal component, quantified the perceived impact of each policing strategy on the reduction of specific criminal demands. Its construction was based on the collective insights of 354 public security specialists, whose responses were systematically captured and standardised using a five-point Likert scale. Subsequently, the Calculation of Veto, Preference, and Indifference Thresholds for each criterion was performed using adapted general formulae, with predefined coefficients for q (0.03), p (0.08), and v (0.2). The subsequent stages involved the Obtainment of the Credibility Matrix and the Obtainment of the Dominance Matrix, both generated through the J-Electre software. Finally, the Obtainment of the Final Ranking was achieved by thoroughly exploring the outranking relationship, which involved constructing two partial classifications (

Z_{1}

and

Z_{2}

) whose intersection (

Z = Z_{1} \cap Z_{2}

) yielded the definitive ranking of strategies.

The empirical findings of the study underscore a critical insight: criminal demands exhibit significant heterogeneity across different geographical areas. Specifically, it was identified that 40% of crimes in AISP 5 (Downtown Rio de Janeiro) showed no correlation with those in AISP 19 (Copacabana, Rio de Janeiro), and conversely, 33% of crimes in AISP 19 were not observed in AISP 5. This heterogeneity directly informed the subsequent phase of the methodology, which sought to identify the most appropriate policing strategies tailored to these distinct demands. By employing the ELECTRE method, eight unique scenarios were constructed, demonstrating unequivocally that a specific set of policing strategies is optimally suited for each particular criminal demand. The ranking of policing strategies, as determined by the ELECTRE IV method, consistently placed directed policing strategies (both motorised and preventive) within the top five positions. This compellingly highlights a clear preference among local public security managers for models rooted in problem-oriented policing and focused policing approaches. Conversely, strategies based on random policing consistently occupied the lowest ranks, thereby reinforcing existing literature that attests to their inherent inefficiency. Furthermore, the sensitivity analysis conducted on the ELECTRE IV model confirmed its remarkable stability in the face of alternative suppressions, indicating that the ranking of strategies remains robust and consistent.

The practical implications of the developed methodology are substantial, offering a complementary yet crucial contribution to understanding criminal practices and their nuanced characteristics, drawing upon detailed reports meticulously stored in 190 emergency service databases. The actionable intelligence derived from the precise identification of criminal demands empowers law enforcement decision-makers to thoroughly evaluate and select from the array of available policing strategies those that most effectively align with the specific realities under consideration, ultimately leading to a measurable reduction in local crime rates. Managers can analyse a multitude of scenarios with enhanced clarity and impartiality, facilitating the selection of the optimal strategy, or indeed a synergistic combination of strategies, with the overarching aim of mitigating criminal indices within a given locality.

From a broader societal perspective, a direct correlation can be inferred between the selection of appropriate local crime-fighting strategies and the optimisation of governmental resources allocated to police agencies. This strategic optimisation effectively alleviates pressure on public budgets, reducing the demand for additional funding. Furthermore, by facilitating the choice of more efficacious strategies, the model directly contributes to the reduction of local crime, thereby fostering an enhanced sense of public safety and security. The intrinsic originality of this study resides in its pioneering integration of text mining techniques (specifically LDA) with the ELECTRE method. This synergistic approach enables the precise detection of crime patterns within each locality, drawing directly from crime reports stored in 190 emergency service databases, and, crucially, facilitates the identification and selection of policing strategies specifically customised to address identified criminal demands. This research unequivocally advances the field of applied police intelligence.

Despite the significant contributions of this research, certain limitations exist, which naturally open up avenues for future inquiry. The collected data, while comprehensive for the studied period (January 2013 to December 2016), primarily reflects the social dynamics of the central and southern zones of Rio de Janeiro. This inherent specificity implies that the derived results may not be directly generalisable to areas possessing different socio-demographic or criminal characteristics. Therefore, future research could benefit from applying the developed research instrument to samples from diverse international contexts to facilitate cross-cultural comparative analyses of the findings.

Further research could also focus on refining the impact matrix. This could involve replicating the study, soliciting perceptions from public security specialists in other states and countries, regarding the impact of public security strategies on specific criminal scenarios. Such an endeavor would serve to consolidate and enhance the robustness of the existing impact matrix. Additionally, the crucial aspect of validating topics in the identification of criminal demands presents a significant area for future work. Expanding the application of validation instruments with a broader spectrum of specialists could facilitate the development of algorithms capable of automating topic validation. Moreover, conducting field experiments to validate the selection of policing strategies under specific scenarios, potentially incorporating advanced geo-referencing tools and linear programming techniques for optimal resource allocation, represents a promising avenue for empirical validation. Finally, a comparative analysis of the results obtained using the ELECTRE I and ELECTRE IV multi-criteria methods with other established multi-criteria methods, such as AHP or ELECTRE TRI, could yield valuable insights. Likewise, investigating the consistency of solutions by comparing the perceptions of specialists within their respective police units against the proposed model warrants further exploration.

Video Upload Options

1. Introduction

2. Research and Findings