02 / Research topics
Potential PhD Research Topics in Data Analytics
Explore suggested topics and the methods, tools and expertise that could support each investigation.
1
Causal and Federated Analytics for Privacy Preserving Healthcare Decision Support
Although healthcare models achieve impressive prediction accuracy, they often cannot establish treatment effects, share sensitive data, or generalize across hospitals. This project would develop federated causal learning methods that combine heterogeneous institutional data while protecting patient privacy. Novel contributions could address confounding, missingness, distribution shifts, and fairness across demographic groups. Objectives include formal identification, scalable algorithms, privacy guarantees, interpretable individualized treatment recommendations, and prospective evaluation using multi-site clinical data.
Applications and use cases
2
Foundation Models for Multimodal Scientific Data Discovery and Hypothesis Generation
As scientific datasets become multimodal and enormous, researchers struggle to connect text, images, graphs, experiments, and sensor observations. This proposal would create domain-adapted foundation models capable of retrieving evidence, integrating modalities, and generating testable scientific hypotheses. The gap lies in limited scientific grounding, weak uncertainty calibration, and poor reproducibility of generated claims. Objectives should include efficient pretraining, knowledge-augmented reasoning, human verification workflows, benchmark creation, and laboratory or simulation-based validation.
Applications and use cases
3
Trustworthy Explainable Analytics for High Stakes Automated Decision Systems
Beyond predictive performance, high-stakes organizations need systems that are understandable, contestable, robust, and fair. Existing explanation techniques can be unstable, misleading, or disconnected from decision makers’ reasoning. This research would develop evaluation frameworks and explanation methods that combine statistical validity, causal relevance, uncertainty, and user needs. Objectives include defining measurable trustworthiness criteria, detecting explanation failures, designing interactive explanations, and testing whether explanations improve calibrated human decisions without increasing automation bias.
Applications and use cases
4
Real Time Urban Digital Twins for Sustainable Mobility and Infrastructure Management
Urban planners lack integrated, continuously updated analytical environments for testing mobility, emissions, infrastructure, and equity interventions. Current digital twins frequently emphasize visualization rather than reliable prediction, causal simulation, or stakeholder participation. This project would build a real-time urban digital twin that fuses IoT, transport, satellite, and administrative data. Novelty could involve hybrid physics-machine learning models, uncertainty-aware scenario testing, and fairness-sensitive optimization. Evaluation should measure accuracy, resilience, sustainability, usability, and policy relevance.
Applications and use cases
5
Energy Efficient Machine Learning for Sustainable Large Scale Data Analytics
While analytics systems are expanding rapidly, their computational and environmental costs are poorly measured and rarely optimized jointly with accuracy. This research would develop energy-aware learning and data-processing methods spanning model design, hardware utilization, scheduling, and carbon-aware deployment. The gap includes inconsistent measurement standards and limited understanding of accuracy-energy tradeoffs under realistic workloads. Objectives include lifecycle accounting, adaptive resource allocation, efficient architectures, reproducible benchmarks, and validation across cloud and edge environments.
Applications and use cases
6
Robust Data Analytics Under Distribution Shifts and Adversarial Environmental Changes
Real-world analytical models degrade when populations, sensors, policies, or environments change, yet many systems assume stationary data. This project would develop adaptive analytics that detect shifts, separate benign change from attacks, and update safely with limited labels. Its novelty may combine conformal prediction, causal invariance, robust optimization, and continual learning. Objectives include shift diagnostics, uncertainty-aware adaptation, adversarial resilience, selective prediction, and evaluation across temporal, geographic, and operationally changing datasets.
Applications and use cases
7
Privacy Enhancing Analytics for Cross Organizational Data Collaboration
Organizations could solve major societal and commercial problems by combining data, but legal, competitive, and ethical barriers prevent direct sharing. This research would compare and integrate privacy-enhancing technologies for useful, auditable collaborative analytics. The gap concerns practical tradeoffs among privacy, accuracy, scalability, interoperability, and governance. Objectives include developing hybrid protocols, quantifying disclosure risk, designing consent and access controls, establishing utility benchmarks, and demonstrating collaboration in finance, healthcare, or public-sector settings.
Applications and use cases
8
Graph Analytics for Detecting Hidden Networks in Complex Socio Technical Systems
Many threats and opportunities arise from relationships rather than isolated records, but conventional analytics often miss evolving, indirect, and multi-layer connections. This project would develop temporal, heterogeneous graph methods for discovering hidden structures while controlling false discoveries and protecting sensitive identities. Novelty could combine graph representation learning with causal discovery, anomaly detection, and expert feedback. Objectives include interpretable network inference, uncertainty estimation, scalable algorithms, and validation on real socio-technical datasets.
Applications and use cases
9
Human Centered Analytics for Collaborative Knowledge Work and Organizational Decision Making
Organizations increasingly deploy analytics into teams, yet research still prioritizes model accuracy over coordination, expertise, power, and accountability. This proposal would study how humans and analytical agents jointly frame problems, interpret evidence, and make decisions. The gap is limited understanding of when automation augments versus displaces expertise. Objectives include designing collaborative interfaces, modeling disagreement and uncertainty, measuring decision quality and inclusion, and conducting longitudinal field studies across knowledge-intensive organizations.
Applications and use cases
10
Causal Analytics for Climate Adaptation and Equitable Environmental Policy Design
Climate impacts are unevenly distributed, while policymakers need evidence about which adaptation interventions work for whom and under what conditions. Existing studies often rely on correlational associations, coarse spatial data, or evaluations that overlook vulnerable communities. This research would combine causal analytics, remote sensing, administrative records, and community knowledge to estimate intervention effects. Objectives include identifying heterogeneous impacts, addressing spatial confounding, quantifying uncertainty, and producing equitable policy recommendations tested through natural experiments or pilot programs.
Applications and use cases