Members’ Publication

Road traffic accidents remain a significant public safety issue in Bangladesh, causing high mortality rates and eco nomic losses.With high death and economic tolls. While measures have been taken to make roads safer, the rise in the number of cars on the road coupled with the lack of proper infrastructure is aiding accidents.The aim of this study is to predict the severity of road traffic accident based on machine learning methods namely LightGBM and SHAP in Bangladesh. The purpose of this study is to find out the number and type of factors which different low and normal severity incidents from fatal incidents have, and to find out how the identified factors can be used for traffic safety improvement. This study has employed a data set on the occurrence of road traffic accidents using FIR data from three districts of Rajshahi, Natore and Pabna in Bangladesh. Data were carefully preprocessed, with missing values addressed and data normalized. The LightGBM model was applied to classify the severity of accidents and SHAP was applied to understand how the individual features contributed to the model’s prediction. The LightGBM model shows great performance, with an AUC of 0.87 for fatal and low/normal severity classes. Type of collision, vehicle type, vulnerable road users and environmental conditions turned out to be the most prominent factors affecting the severity.Overall, the study highlights the potential of machine learning in predictive modeling for improving road safety. These findings can help policy makers prioritize the interventions that are needed for safety, by focusing on the factors that are most likely to lead to death, thus minimizing the number of lives lost on the roads in Bangladesh.

Life, livelihood, nature, and the onset of extreme climate change continue to pose challenges for the coastal residents of Dacope Upazila in Khulna district, Bangladesh, almost daily. As Dacope is one of the key areas affected by climate change, it is highly vulnerable to these changes. Recognizing the hardships faced by the community and the need for targeted support, this study evaluates the impacts of climate change on various livelihood assets using the Livelihood Vulnerability Index (LVI), LVI-IPCC, and Livelihood Effect Index (LEI) methods. Data were collected through a semi-structured household questionnaire survey covering nine unions within Dacope Upazila, with a sample of 400 households. Secondary data was obtained from the NASA Power website. Results show that Sutarkhali Union is the most vulnerable, with an LVI score of 0.584. Both the LVI-IPCC and LEI approaches ranked Sutarkhali as the most vulnerable. This study provides insight into living conditions and the factors that make the livelihoods of Dacope Upazila’s residents vulnerable to climate change. Consequently, it offers valuable information for policymakers and development practitioners to design effective interventions for building resilience in coastal communities.

Despite the growing progression of regional economic integration as a driving factor for national development, the spatial dynamics of industrial interdependence in Bangladesh's southern coastal zone remain insufficiently theorized and empirically underexplored. The research addresses this gap by examining the functional correlation, dependable characteristics, and structural linkages among industries across four strategically significant districts (Khulna, Satkhira, Bagerhat, and Narail) through an integrated spatial-analytical framework. Employing a composite weighted index for formal regionalization alongside Reilly's Law of Retail Gravitation, the research delineates influence areas (IA) and unserved areas (UA) at the upazila level to assess intra-regional economic reach and service accessibility differentials. Input-output interdependence is contextualized within a GIS-based spatial representation, while Local Indicators of Spatial Association (LISA) and Moran's I statistic are applied to detect local clustering patterns, spatial outliers, and distributional asymmetries in industrial concentration. The findings reveal considerable spatial heterogeneity: Koyra (Khulna) and Shyamnagar (Satkhira) exhibit the highest influence areas, whereas Daulatpur registers the lowest, reflecting structural deficits in transport connectivity and economic infrastructure. At the district level, Narail records the lowest unserved area share (0.96%), attributable in part to its favourable positioning along the Padma Bridge corridor. LISA analysis identifies the absence of significant high-high clusters, indicating a fragmented rather than cohesive pattern of industrial development, with Khulna emerging as a high-influence, high-unserved-area outlier characterized by disproportionately low industrial concentration relative to its functional importance. These spatial imbalances are further explained through a multi-factor analysis encompassing transport networks, vocational infrastructure, agricultural land use, and market density. The research concludes that targeted policy interventions, including upazila-scale industrial decentralization, infrastructure investment, and skills development, are necessary to foster inter-regional industrial interdependence and mitigate spatial disparities in southern Bangladesh.

Cyclones, storm surges, sea-level rise, tidal flooding, and limited adaptive capacity due to high population pressure and flat terrain make coastal Bangladesh highly vulnerable. This study assesses long-term climatic variability and household-level social vulnerability (SV) in Dacope Upazila, Khulna District, Bangladesh. Primary data were collected through semistructured household surveys using random sampling, while secondary climate data (1981–2021) were obtained from the NASA POWER database. The Mann–Kendall test was applied to evaluate the statistical significance of trends in rainfall and temperature, and Sen's slope method was used to estimate trend magnitude. Results indicate a downward trend in precipitation for January and February, with Z values of −0.096 and −0.042, respectively. Annual temperature also shows a decreasing trend (Z = −0.422). The rainfall anomaly index reveals that since 2015, values exceeding 1.3 indicate annual precipitation shifted from dry to extremely humid conditions. Future projections using global climate models suggest that maximum temperature may increase by 1–2 °C. SV was assessed using principal component analysis based on 28 indicators. The SV index score (+9.772) indicates high societal sensitivity to climate change. These findings can support policymakers in developing effective climate adaptation and disaster risk reduction strategies.

asd

asd

Cities worldwide are quickly developing due to population increase, economic development, and infrastructural improvements in the 21st century. MRT may transform urban mobility and the landscape by providing reliable, efficient, and sustainable transportation. This study seeks to identify the impact of newly established metro stations on the urban landscape and property values. The study focuses on Uttara South and North, two sizable metro stations in the capital city of Dhaka, Bangladesh. Shannon’s Entropy was applied to measure land-use dispersion and detect spontaneous development. Mann-Kendall test analysis was conducted to identify factors contributing to expansion. The Hedonic Pricing Model (HPM) was also employed to assess the impact of proximity to metro stations on residential property values. Additionally, a SWOT analysis was performed. Data was gathered through a questionnaire survey, Key Informant Interviews, and the Landsat 8 satellite. Results show moderate dispersion in entropy values, suggesting growth is structured. The HPM identified eight attributes contributing to the increase in property prices. With p-values below 0.05, each of the eight attributes showed statistical significance. Uttara North Metro Station’s entropy value rose from 1.20 in 2020 to 1.23 in 2024. Likewise, the entropy value of Uttara South Metro Station increased from 1.12 in 2020 to 1.18 in 2024. The Kendall tau value (1) shows upward trends, indicating that increasing population and housing demand are among the significant factors contributing to expansion. This research advocates affordable housing and inclusionary zoning rules to share Transit-Oriented Development (TOD) benefits evenly across socioeconomic levels to foster socio-economic diversity.

Transport mode choice is a crucial determinant of urban mobility patterns, influencing infrastructure planning, environmental sustainability, and social equity. Although it is important, there is still an unfulfilled gap in the use of socio-demographic, behavioral and perception-based factors to accurately predict choice of transport mode especially in the developing countries. The proposed study shall fill this gap by applying a deep learning model to make predictions of transport mode choice using a multidimensional survey data. The primary objective of this research was to investigate the potential of a Gated Recurrent Unit (GRU) neural network for classifying transport mode choice using a variety of socio-demographic, economic, and travel behavior features. The dataset includes 348 respondents, capturing responses related to seven transport modes: Bike, Bus, CNG, Car, Cycle, Easy bike, and Rickshaw. Key attributes in the dataset include gender, age, occupation, income, travel objectives, and safety perceptions related to different transport modes. The preprocessing of the data included the following: one-hot encoding of features of character type, normalization of numeric data, and label encoding of target variable. The GRU-based model was trained and tested based on stratified 80:20 train-test split and tested with reference to confusion matrices, ROC, and precision-recall. The accuracy of the model has a value of 79%, ensuring very good results on modes such as CNG (precision = 1.00, recall = 1.00) and Easy bike (precision = 0.73, recall = 0.88). This class-wise analysis raised the issue of less frequent mode prediction such as Cycle and Car modes which were having a lower recall. The AUC scores had micro- and macro-averages of 0.943 and 0.918 respectively. These findings reveal the ability of deep learning models, especially GRUs to be used reliably to predict transport mode preferences using a diversified group of socio-demographic and behavioral variables. This predictive modeling would provide urban designers and policymakers with the information needed to design more specifically focused interventions to promote sustainable transport networks, increase modal share, and reduce congestion in urban areas.