RESEARCH & TEACHING
My academic work focuses on recurring weaknesses in AI and data practice: poorly understood datasets, unexamined performance measures and processes that jump to modelling before establishing evidence quality.
The purpose is not to add academic language to delivery. It is to make technical claims more testable, comparable and defensible.
Academic qualification: Associate Professor qualification in Computer Engineering; PhD from Middle East Technical University.
RESEARCH THEMES
Feature-frequency distributions, source differences, dataset comparison and systematic profiling—particularly for high-dimensional binary-feature datasets.
Practical relevance: AI readiness, source comparison, migration risk, labelling quality and fitness-for-purpose decisions.
BenchMetrics and related work examine whether Accuracy, F1 and other measures behave robustly enough to support the intended decision.
Practical relevance: evaluation design, metric rationale, model comparison and defensible acceptance criteria.
Process structures that treat dataset-quality analysis as a control gateway, rather than an optional preparation task.
Practical relevance: stage gates, evidence requirements and separation of data, model and system quality.
Malware, mobile security, taxonomies, secure information sharing, awareness and strategic security thinking.
Practical relevance: threat reasoning, security architecture, governance and education.
SELECTED PEER-REVIEWED WORK
A systematic method and meta-metrics for testing the robustness of classification-performance measures.
View DOI ↗Feature-space distribution fitting to understand and compare datasets before model development.
View DOI ↗A systematic knowledge representation and periodic-table-style structure for a large performance-measure landscape.
View DOI ↗Systematic benchmarking of probabilistic error and loss instruments across designed evaluation cases.
View DOI ↗A performance indicator that exposes cases where reported Accuracy is confounded by a simple baseline or class distribution.
View DOI ↗Techniques and dimensions for profiling large ML datasets, illustrated through Android mobile-malware data.
View DOI ↗TOOLS · ASSETS · PATENTS
Research method, experimentation material, metric-space data and an open-source API for examining performance-measure robustness.
Explore BenchMetrics ↗A compact research and education tool for calculating and visualising binary-classification performance instruments.
View paper ↗Named inventor in a connected route-guidance family covering near-real-time guidance and local/global walkable-path generation.
Representative public record ↗TEACHING
The courses are in Turkish and progress from foundations to advanced analysis, malware and enterprise security governance.
RESEARCH → ARCHITECTURE VALUE
RESEARCH & ADVISORY OPPORTUNITIES
Relevant work includes independent AI or data-quality evaluation, technical review, expert panels, proposal evaluation, advisory boards and workshops tied to a real architecture decision.