Research Overview

Welcome to the Applied Machine Learning Lab. Led by Prof. Dr. Rafet Sifa, our lab develops hybrid, interpretable, and resource-aware learning systems that combine statistical machine learning with knowledge-driven methods. We focus on robust, trustworthy, and deployable AI that solves real-world problems across medical and life science informatics, text mining, and behavioral analytics. Our mission is to translate cutting-edge research into practical solutions that improve decision-making, efficiency, and impact.

In medical and life science informatics, we work with multimodal clinical and biomedical data—including imaging, sensor streams, genomics, and scientific literature—while addressing privacy, bias, and regulatory constraints. We design hybrid models that integrate domain knowledge, ontologies, and causal priors with deep learning to deliver interpretable decision support for diagnostics, risk stratification, digital phenotyping, and precision medicine. Our systems emphasize resource-aware learning for edge and low-resource settings, privacy-preserving training via federated and differential privacy techniques, and human-in-the-loop validation to ensure safety and accountability. By fusing heterogeneous signals and grounding models in evidence, we accelerate triage, synthesize literature for guideline support, and make clinical AI more reliable and transparent.

In text mining, we build robust NLP for high-stakes domains, focusing on specialized, long-form, multilingual, and low-resource corpora across scientific, clinical, financial, and legal contexts. We develop representation learning and retrieval-augmented approaches that keep models factual and auditable, with parameter-efficient adaptation for scalability. Our work spans information extraction, summarization, and question answering, supported by rigorous data curation, uncertainty estimation, and evaluation frameworks to reduce hallucinations and ensure traceability. These systems enable accurate document parsing, compliance assistance, and scalable evidence mining, creating trustworthy text analytics that integrate seamlessly into real-world workflows.

In behavioral analytics, we model human behavior, personality traits, and decision-making through digital interaction signals from gaming, web and app usage, and sensors. We use controlled gaming environments to derive reproducible behavioral markers and then generalize to real-world settings through hybrid models that combine psychological theory with graph, sequential, and causal structures. Our research addresses context-dependence, individual variability, transferability, and ethics by embedding fairness, privacy-by-design, and transparent reporting throughout the pipeline. This work supports personalization, early warning and intervention for engagement and well-being, and deeper insights into human-computer interaction and digital health.

Across all areas, we prioritize practical, safe, and interpretable AI. Our systems are co-designed with domain experts, validated in real deployments, and, where possible, released with open-source components. We bridge technology and everyday challenges to deliver measurable, responsible impact.