Ph.D. Theses
Permanent URI for this collection
Browse
Browsing Ph.D. Theses by Author "Badur, Bertan Yılmaz."
Now showing 1 - 3 of 3
Results Per Page
Sort Options
Item A 360° customer lifetime value prediction method using machine learning for multi category e-commerce companies(Thesis (Ph.D.) - Bogazici University. Institute for Graduate Studies in the Social Sciences, 2023., 2023) Yılmaz, Gülşah.; Badur, Bertan Yılmaz.; Mardikyan, Sona Kunuzyan.The Customer Lifetime Value (CLV) prediction methods that are used by e-commerce companies are mainly focusing on a specific group of customers with multiple transaction data and therefore remain unable to value the one-time purchasers. As a result, the management of these companies is incapable of valuing all customers and the overall company. Thus, each type of user needs a different way to predict CLV. In this thesis, we intend to develop a novel 360° holistic technique for the prediction and use of CLV models in the marketing management of multi-category e-commerce companies to enhance their strategic decision-making. The research compared the proposed framework which was constructed with several outputs (CLV, DPC, and TAS) with other ML models to evaluate the new variables created based on relationship marketing theory (RMT) and to demonstrate the best model which is more appropriate for multi category e-commerce companies' usage. To make this result useful, we created customer clusters that enable management to separate end-users according to the three outputs. Finally, Shapley values obtained using explainable artificial intelligence (XAI) are then utilized to understand the DNN's findings. The results showed that using XAI shows which factors are more crucial to the results. Overall, the proposed model helps marketing management teams in planning their operations efficiently by differentially allocating their resources to specific types of customers based on their profitability which provides more strategic decision-making.Item Developing a dynamic predictive policing system(Thesis (Ph.D.) - Bogazici University. Institute for Graduate Studies in the Social Sciences, 2022., 2022.) Hakyemez, Tuğrul Cabir.; Badur, Bertan Yılmaz.The retrospective predictive policing techniques are atheoretical and therefore remain incapable of sensing the changing crime risk across the streets. In this study, we aim to develop a dynamic predictive policing system that capitalizes on theory-based risk indicators. The sample includes all the theft and robbery incidents in Chicago between 2014-2019. In the first step, pipelining bivariate network K analysis and segmented regression, we introduce novel distance-aware risk functions that operationalize spatiotemporal crime risk around the selected urban features (i.e., bus stop, fast food restaurant, gas station, grocery store, pub). In the second step, we develop various network-based predictive policing methods using graph-based deep learning algorithms (i.e., GraphWavenet, Spatiotemporal Graph Convolutional Networks). These methods generate weekly and intraday hotspot predictions. We complement these methods with various theory-based risk indicators including a risk score devised from the novel risk functions, 311 calls, park events, and cooccurring crime incidents. The results showcase that crime risk around urban features varies across space, time, and crime types. Furthermore, this risk is found to be significantly correlated with the regional socioeconomic characteristics. Another important result shows that incorporating theory based indicators improved the performance of the retrospective methods up to 68%. Amongst the algorithms, GraphWavenet is found to outperform its counterparts in the majority of the prediction models with an accuracy as high as 80%. The proposed system helps law enforcement agents in planning their operations efficiently by pinpointing the micro geographical units with relatively higher risks in the next time step.Item Essays in learning representations of complex networks(Thesis (Ph.D.) - Bogazici University. Institute for Graduate Studies in the Social Sciences, 2021., 2021.) Gürsoy, Furkan.; Badur, Bertan Yılmaz.This thesis contains three essays in learning representation of complex networks, the first two of which develop new methods and the third utilizes these methods in a real-world application. The first essay provides methods for extracting underlying signed network backbones from intrinsically dense weighted networks. Utilizing a null model based on statistical techniques, we propose significance and vigor filters that enable inferring edge signs and weights. Empirical analysis on four real-world networks reveals that the proposed filters extract meaningful and sparse signed backbones that exhibit characteristics typically associated with signed networks while respecting the multiscale nature of the network. The second essay deals with the misalignment problem in dynamic representation learning. We provide the first formal definitions of alignment and stability, propose novel metrics for measuring them, and show their suitability through a set of synthetic and real-world experiments. We show that, by ensuring alignment, the performance of dynamic network inference tasks improves by a remarkable amount. The third essay applies the novel methods developed in the first two essays as well as other methods from the network analysis literature to investigate the structure and dynamics of internal migration in Turkey. In addition to providing unique and specific insights, we find that most migration links are geographically bounded with exceptions of cities with large economic activity, migration takes place in well-defined routes, counter-streams develop for major migration streams, and the migration system is largely stable over time; which are generally in line with classical migration laws.