FATİH SAHIN

Dr. FATİH SAHIN

orcid

Research Focus & Metrics

Academic Career Length

4Years (2023-Present)

(Calculated once Scilit publication data is synced: current year − year of first publication + 1)

Representative Publications

View All Publications
16 January 2023 in Applied Sciences
Verified
Journal Article
Open access
PDF file
GRGhulam Rasool
Ghulam Rasool
YHYasir Hussain
Yasir Hussain
TUTariq Umer
Tariq Umer
Jawad RasheedJawad Rasheed
Jawad Rasheed
Sook Fern YeoSook Fern Yeo
Sook Fern Yeo
FSFATİH SAHIN
FATİH SAHIN
Software design patterns have a proven impact on the quality of software applications and the development process of an application. The success of design patterns in the software industry has attracted mobile game developers and researchers to apply patterns in the context of mobile games. Researchers have already proposed different frameworks and design patterns, but they are not truly beneficial for game developers. The high-level taxonomies can be adjuvant while proposing useful design patterns. The existing taxonomies for mobile games do not consider different parts of a game that outline top-level structure. In this paper, we propose a new taxonomy that emphasizes the top-level structure for identifying new design patterns for mobile games. We propose five novel generic design patterns that might be applied to the development of mobile games and other software applications. The presented design patterns are, in a true sense, programming patterns that outline top-level generic classes and interfaces, and that could be the basis for the development of new games. We developed four demo games by using these patterns for the realization of taxonomy and design patterns.

25 October 2023 in Dicle Üniversitesi Mühendislik Fakültesi Mühendislik Dergisi
Verified
Journal Article
Open access
PDF file
FSFATİH SAHIN
FATİH SAHIN
GTGökalp Tulum
Gökalp Tulum
ŞKŞeyma Karaca
Şeyma Karaca
Sağlık sektöründe hastalıkların teşhisi için yapay zekânın alt dallarından olan makine öğrenmesi oldukça yaygın kullanılmaktadır. Çalışmada anne sağlığı riski üzerine bir veri seti kullanılarak hamilelikte risk üzerine sınıflandırma çalışması yapılması amaçlanmıştır. Çalışmada, makine öğrenmesi algoritmalarından lineer regresyon, destek vektör makineleri, karar ağacı algoritması, rastgele orman algoritması, çok katmanlı algılayıcı, naive bayes sınıflandırıcısı, k-en yakın komşu algoritması ve XGBoost sınıflandırıcısı kullanılmıştır. Aynı zamanda her bir algoritma için temel bileşenler analizi (PCA) ve lineer diskriminant analizi (LDA) uygulanmış olup oluşturulan modellerin doğruluk oranlarına bakılarak tahminde bulunulmuştur. Yapılan tahmin sonucunda en yüksek doğruluk oranı %84 ile rastgele orman algoritmasından, PCA dönüşümü kullanılarak yapılan tahmin sonucuna göre en yüksek doğruluk oranı %82 ile rastgele orman algoritması ve K-en yakın komşu algoritmasından ve LDA dönüşümü kullanılarak yapılan tahmin sonucuna göre ise en yüksek doğruluk oranı %85 ile karar ağacı algoritması ve K-en yakın komşu algoritmasından elde edilmiştir. Sınıflandırma işleminde LDA dönüşümünün daha yüksek sonuç elde ettiği görülmektedir.

3 July 2026 in Preprints.org
Verified
Preprint
Open access
PDF file
FSFATİH SAHIN
FATİH SAHIN
Cross-organization cyber defense must reconcile collaborative learning with privacy and adversarial robustness — yet standard federated learning ships full gradient tensors, leaking sensitive posture and inviting Byzantine manipulation. We present FedMARL-LTI, a federated multi-agent reinforcement learning framework whose architecture answers both pressures with a single decision: organizations share neither raw data nor model weights, only differentially-private 768-dimensional semantic threat embeddings. The contribution is fourfold. (1) Semantic Abstraction (SA) channel: per organization, each round, the local gradient is summarized by an LLM, projected to a 768-dim embedding, L2-clipped, and Gaussian-noised before any numeric quantity leaves the host. The bottleneck reduces the per-element noise scale from O(√(d_model )) to O(√m) with m=768≪dmodel≈3×105. (2) Formal privacy analysis: the SA+DP cascade satisfies (ε,δ)-DP and bounds per-round mutual-information leakage by min{Ttoklog2V, m⁄2 log2 (1+C2/(mσ2))} (Theorem 1), with Rényi composition over T federation rounds (Theorem 2). (3) Byzantine-resilient ClippedClustering aggregator combining L2 clipping with cosine-similarity clustering. (4) Hierarchical MARL policy with threat-profile-aware LLM-IRR reward shaping, wired end-to-end and disclosed honestly (the LLM call is currently stubbed with a deterministic projection for reproducibility). We evaluate on CybORG CAGE-4 with n=5 organizations, 30 federation rounds × 5 episodes × 100 steps per round. The SA channel adds statistical-zero utility cost vs. no-privacy baseline: SA-only Δreward = -0.66 (t=+0.31, NS), dual SA + Weight-DP Δreward = +1.90 (t=-0.71, NS), all N=5 seeds, all |t|< 1.3. A controlled signal/noise probe confirms a 19.58× improvement of SA over Weight-DP at fixed DP budget — matching the predicted √(d/m)≈19.8. Under Byzantine sign_flip at 30% (N=15), ClippedClustering is directionally strongest (F1=0.025 vs FedAvg 0.020, Krum 0.016) but the edge is not statistically significant (CC vs Krum t=+1.59, p=0.15, d=+0.58; the earlier N=5 “3.4×” gap was small-sample optimism, §5.2); its decisive Byzantine win is the harsher random_noise attack, where FedAvg diverges to NaN and Krum collapses to 0.002 while ClippedClustering survives at 0.020 (§5.7, Cohen’s d=+3.77). The cooperative-PPO family (MAPPO, IPPO) outperforms value/actor-critic (QMIX, MADDPG) by ≈20 reward units, p< 0.001. All host-level F1 values stay below 0.05 at the 15K-step training horizon used here; the relative claims of the paper (privacy zero-cost, ClippedClustering’s decisive Byzantine win on the harshest attacks per §5.7, cooperative-PPO dominance) are unaffected by this scope. A 200K-step long-horizon replication (§6.3 L1) lifts F1 above the 15K plateau (to ≈0.044, N=5) — confirming that horizon, not the privacy/Byzantine machinery, gates absolute accuracy — but a finer 60-checkpoint run shows the climb is volatile and non-monotonic and does not reach deployment-grade, an honest stability-not-compute limitation. We release all 141 raw run JSON outputs (Phases 1–3, the L4 backend comparison, and the algorithm/aggregator baselines), the figures, and analysis scripts for replication.

20 August 2026 in Preprints.org
Verified
Preprint
Open access
PDF file
FSFATİH SAHIN
FATİH SAHIN
NMNecibe Sare Mert
Necibe Sare Mert
Security operations centres (SOCs) face two simultaneous mandates: keep the network traffic they analyse private, and make their AI-based detectors explainable. These goals are usually assumed to be in tension, because in image and language models differential privacy (DP) is reported to degrade post-hoc explanations. We present the first systematic audit of explanation faithfulness under formally-accounted (ε,δ)-DP in federated, tabular deep intrusion detection (IDS), and we add a previously-unstudied axis: the interaction between Byzantine-robust aggregation and explanation quality. Across two datasets (ToN_IoT, CICIDS2017), two architectures (an MLP and a compact tabular transformer), three attribution methods (Integrated Gradients, GradientSHAP, attention rollout), privacy budgets ε ∈ {8, 4, 1} (ε ∈ {4, 1} for the transformer), and heterogeneity levels, and using paired Wilcoxon tests with Cliff’s δ effect sizes, we find — contrary to our own anchor hypothesis and to the image/NLP literature — that client-side DP-SGD does not degrade explanation faithfulness. Measured by comprehensiveness and the AOPC MoRF−LeRF separation it significantly increases self-faithfulness (large δ, all seeds), an effect that attenuates at the strongest budget and survives a confidence-ceiling (f0) normalization. A pre-registered capacity-matched control attributes this faithfulness gain largely to DP’s utility cost — non-DP models throttled to the same accuracy by three independent mechanisms reproduce it — leaving only a small, clipping-driven residual in the centralized setting. DP’s real cost falls on utility and, most sharply, on rare-attack detection (mitm F1 0.65→0.05). Per-class utility and faithfulness decouple: well-supported classes hold their F1 while their attribution gap widens. A strong model-poisoning attack collapses FedAvg in both utility (0.83→0.20) and faithfulness (gap 0.58→0.21), whereas median, trimmed-mean and multi-Krum preserve both — though median/Krum cost some faithfulness when no attack is present, making trimmed-mean the best all-round trade-off. Among attribution methods, attention rollout is the most fragile. Practically: for explainable federated IDS, gradient/Shapley attributions remain trustworthy under DP, and trimmed-mean aggregation buys robustness at little explanation cost.

20 July 2026 in Preprints.org
Verified
Preprint
Open access
PDF file
FSFATİH SAHIN
FATİH SAHIN
NMNecibe Sare Mert
Necibe Sare Mert
Automated alert triage can reduce Security Operations Center (SOC) workload, yet the validation-tuned thresholds deployed systems rely on carry no finite-sample control of their operational error rates and degrade unpredictably under distribution shift. We present a model-agnostic dual-threshold conformal deferral architecture: high-score alerts are auto-escalated under finite-sample marginal class-conditional control of the benign-escalation probability (budget α), low-score alerts are auto-closed under matching control of the threat-miss probability (budget β), and the rest are deferred to an analyst. It needs no retraining, and closes an automatic zone rather than certifying what the calibration data cannot support. We evaluate it on a reinforcement-learning investigation agent in a simulated SOC and on four classifiers trained on CIC-IDS2017 and tested on CSE-CIC-IDS2018, using stratified 25,000-flow calibration and evaluation samples, with attack-type recall computed over the full 16.2-million-flow corpus. Pooling episodes from ten trained policies across two evaluation datasets, the architecture automated 73.7% of decisions at α = β = 0.01, realizing benign auto-escalation and threat auto-close rates of 0.0099 and 0.0101, and deferring the hardest ~26% of alerts (≈49% threat prevalence). After recalibration on labeled target-domain data, severe cross-dataset degradation appears not as silent error but as sharply reduced certifiable automation, with deferral rising to 79–99% for the most affected classifiers.

17 September 2026 in Preprints.org
Verified
Preprint
Open access
PDF file
FSFATİH SAHIN
FATİH SAHIN
Deep learning models for Network Intrusion Detection Systems (NIDS) can detect attacks with high accuracy, but they work as black boxes. Security analysts cannot see why a model flags a certain flow as malicious. This limits trust and makes it harder to deploy these systems in practice. We propose HiT-IDS (Hierarchical Transformer for Intrusion Detection with Intrinsic Explainability), an architecture that detects attacks while also explaining its own decisions. HiT-IDS groups the 72 network flow features into four categories—temporal, volumetric, protocol, and behavioral—and uses separate intra-group Transformer encoders for each group, with per-feature tokenization. An inter-group Transformer then combines the group-level outputs. A two-stage cascade first separates benign from attack traffic (binary F1 > 0.98), then classifies attacks into 12 types. We test HiT-IDS on CIC-IDS2017 with 5-fold stratified cross-validation. Adaptive SMOTE balancing is applied inside each fold to avoid data leakage, with class-specific oversampling targets that prevent noisy synthetic generation for extreme minority classes. A two-phase training pipeline first optimises classification performance (Phase 1), then refines attention patterns through entropy regularisation (Phase 2). HiT-IDS reaches a macro-F1 of 0.9891 ± 0.0011 across 13 classes. We compare it against Random Forest (0.9989), XGBoost (0.9990), BiLSTM (0.9966), and Vanilla Transformer (0.9941). Paired t-tests, Cohen’s d, and 95% confidence intervals confirm that all differences are statistically significant. A primary strength of HiT-IDS is its built-in explainability. Hierarchical attention rollout shows that protocol-level features (importance: 0.306) matter most for attack detection. We employ a Spearman-based consistency check between intrinsic and post-hoc explanations, denoted XCS, to verify that attention-based feature importance agrees with SHAP values. We supplement attention rollout with input-gradient sensitivity (attention×gradient) to capture both information routing and feature sensitivity. XCS yields a mean Spearman ρ of 0.498 ± 0.103 (p < 0.05 for 99.2% of 500 samples, median ρ = 0.503). A cross-model XAI comparison confirms that HiT-IDS learns the same discriminative features as all baselines (per-sample SHAP agreement ρ = 0.65–0.75). Ablation studies confirm that each architectural component—hierarchical grouping, cascade detection, Phase 2 attention regularisation, and GELU activation—contributes to the final performance. HiT-IDS provides per-sample explanations at 0.55 ms/sample through attention rollout—approximately 870× faster than the average post-hoc SHAP cost in our setup—making it one of the few models in our comparison that combines real-time inference with per-sample hierarchical explanations.