Unpacking the Complexities of Armed Conflict Fatalities in Bangladesh: A Data-driven Study of Factors, Actors, and Spatial Patterns

Submission Deadline-30th April 2024
April 2024 Issue : Publication Fee: 30$ USD Submit Now
Submission Deadline-20th April 2024
Special Issue of Education: Publication Fee: 30$ USD Submit Now

Unpacking the Complexities of Armed Conflict Fatalities in Bangladesh: A Data-driven Study of Factors, Actors, and Spatial Patterns

 Sondip Poul Singha, Md. Shamiul Islam, Susmoy Bless Singh, Julkar Naeem
Bangladesh University of Business and Technology, Dhaka, Bangladesh
DOI: https://doi.org/10.51584/IJRIAS.2023.8724
Received: 04 July 2023; Revised: 10 July 2023; Accepted: 14 July 2023; Published: 15 August 2023

Abstract—Bangladesh, a developing country, faces various challenges that hinder its progress. One significant issue is the high crime rate, along with its lower resilience score on the global peace scale compared with other Asian countries. This study investigates the underlying factors that contribute to armed conflict in Bangladesh. Key questions were explored, such as identifying the regions most affected by conflicts, understanding the involvement of different actors in these regions and events, and developing predictive models for fatality rates and future crime based on various related attributes. To address these objectives, machine learning algorithms and clustering techniques were employed in this research. The ACLED[1] Bangladesh dataset, encompassing conflict events from 2010 to 2021, was analyzed to obtain valuable insights. Clustering techniques, specifically k-means and hierarchical clustering, were applied to classify Bangladeshi Divisions and Cities based on shared characteristics. Furthermore, this study investigates the events and actors associated with each cluster to identify hidden factors.

IJRISS Call for paper

Machine learning algorithms are utilized to predict fatality rates by employing various techniques, such as pre-trained models and discretization methods. Finally, the focus shifts towards predicting future crimes by utilizing the Random Forest algorithm, which achieved a 97% accuracy rate. The results of this study demonstrated promising outcomes, with high R2 scores which is Goodness of fit measure, indicating a 99% satisfaction level for predicting fatalities. Overall, this study highlights the potential of machine learning to understand and mitigate conflicts in Bangladesh. It emphasizes the importance of interdisciplinary approaches and stakeholder engagement in developing context- specific tools for effective conflict analysis and mediation. By leveraging the findings of this study, policymakers and relevant authorities can make informed decisions to address the increasing prevalence of crime and work towards a more peaceful and secure Bangladesh.

Index Terms—Conflict events, Fatalities, Armed conflict, Hierarchical Clustering, Machine learning, Spatial analysis, and Predictive modeling.

I. Introduction

Bangladesh emerged in the second decade of the twenty- first century as a country to be reckoned with, but there have been numerous conflicts. The current population of Bangladesh stands at approximately 169.17 million[2]. Studying Bangladeshi armed conflict is crucial for understanding violent conflicts in South Asia and developing effective conflict prevention and peacebuilding strategies in the region. The ACLED[1] Bangladesh dataset provides crucial information on the traits and dynamics of these conflicts, including the type of incident, location, participants, and death tolls. Understanding these patterns and trends is crucial for developing effective conflict prevention and resolution strategies. In this study, we investigated the ACLED[1] Bangladesh dataset using machine learning methods and clustering approaches to gain insights into the patterns and trends of conflict events in Bangladesh. We specifically want to answer several questions, including how event types and actor participation affect the number of fatalities, conflict-prone regions, and the events with the highest fatality rates. Additionally, we investigated the use of machine learning algorithms to anticipate crime rates for the following year and to model and predict deaths based on the given variables. Key Highlights of the Study are:

• Utilizes machine learning algorithms and clustering techniques to examine the ACLED[1] Bangladesh dataset, providing a unique approach to understanding conflict events and fatalities in the region.
• Offers insights into how event types and actor involvement impact the number of fatalities, conflict-prone regions, and the most fatality-prone events.
• Uses a range of machine learning algorithms, including ensemble learning methods, to predict fatalities and anticipate crime rates for the following year, providing valuable information for law enforcement agencies and policymakers.
• Achieved satisfactory results, indicating the effectiveness of the study’s methodology and approach.