Wasit Journal of Computer and Mathematics Science

This study aims to determine whether attention-based explainable AI (XAI) intrusion detection models
can reliably detect zero-day threats in IoT/IIoT networks while remaining interpretable and deployable at the edge.
We benchmark sequence- and graph-oriented attention architectures (Transformer+Attention, Temporal
CNN+Attention, and GAT-based models) against non-attention deep and classical baselines (e.g., LSTM, Random
Forest, Isolation Forest) on multiple IoT intrusion datasets (Bot-IoT, ToN-IoT, UNSW-NB15, and a CIC-IoT-like
corpus) using complementary zero-day protocols (leave-one-attack-family-out, chronological, and cross-domain
transfer). Models are evaluated for detection, calibration, robustness (domain shift, noise/packet loss, adversarial
feature perturbations), explanation faithfulness/stability (insertion/deletion fidelity, comprehensiveness/sufficiency,
sparsity, consistency), and edge efficiency. Across scenarios Z1–Z4, the Transformer+Attention achieves the
strongest zero-day detection (e.g., Z1 ROC-AUC/PR-AUC/F1 = 0.985/0.962/0.928 and Z4 = 0.892/0.751/0.734),
with consistently better low-false-alarm sensitivity (TPR@0.1% FPR 0.842 → 0.577) and lower error/calibration loss
(EER 0.045 → 0.134, Brier 0.032 → 0.071) than baselines. Under cross-domain stress, it remains best-performing
(Z3 ROC-AUC/PR-AUC 0.914/0.792). Edge optimization preserves performance while improving deployment cost:
latency 12.5 ms → 7.4 ms (INT8) → 6.2 ms (pruned), memory 210 MB → 120 MB → 95 MB, and energy 42.0 mJ
→ 25.5 mJ → 21.8 mJ per inference. Explanation quality is retained or improved after compression (e.g., deletion
fidelity up to 0.851, consistency across seeds up to 0.895). Overall, attention-based XAI IDS models provide a strong
accuracy–robustness–interpretability trade-off for practical zero-day IoT defense, with feasible edge deployment
profiles. We recommend reporting full reproducibility settings, adding attention sanity checks (randomization),
extending calibration reporting (ECE/reliability), and adopting shift-aware validation and human-in-the-loop
workflows for operational trust.

The complex Q-fuzzy set (CQFS) is an innovative concept that was created by expanding the concept
of the complex fuzzy set (CFS) by expanding the universal set X into the Cartesian product between X and the set Q.
In this paper, we investigate this idea in a matrix setting by introducing the concept of complex Q-fuzzy matrices
(CQFM). We discuss some new elementary operations, including the complement of CQFM, the union, and the
intersection between two CQFMs. Moreover, we will show some important algebraic properties that link these
processes in a theoretical framework, as well as we familiarize some numerical examples based on CQFM. Then, we
promote techniques for determining and adjoins as well as develop methods for computing composition functions to
determine the greatest and least eigenvalues of CQFMs. Finally, we improve a multi-step algorithm expending CQFM
to represent a problem-solving mechanism in a real-life scenario.

The heterogeneity of the imaging and cognitive findings in early Alzheimers disease (AD) makes it
difficult to make a definite diagnosis. In this study, an initial hybrid decision-support framework of MRI features data
and Cognitive Questionnaire data are proposed in a combination of fuzzy logic with Support Vector Regression
(SVR). 120 subjects were processed through MRI and resizing, denoising, normalization, feature extraction, and 25-
item cognitive evaluation by methods for memory, attention, orientation (for example: would you recall back the
items that we have asked about within just 5 minutes) and daily activities (What day is today? What month? What
year?) Triangular fuzzy numbers were used to represent cognitive responses, merged with MRI features using a
weighted strategy and processed through a fuzzy inference system and fine-tuned with SVR RMSE-based RBF kernel.
The validation method for performance was 5-fold cross-validation. The framework achieved accuracy of 91.9%,
precision of 92.5%, recall of 91.1%, F1-score of 91.8% and ROC_AUC 0:94 high up performance metrics MSE =
0.085 and R² = 0.92 from internal agreement on predicted and observed fuzzy diagnostic scores in SVR refinement.
These findings were further supported by visual analyses, which demonstrated distinct separation of trends among
healthy, pre-Alzheimer's and Alzheimer's groups with partial overlap between adjacent categories. These findings
indicate that fuzzy-SVR is a feasible preliminary screening model, with good interpretability. But because of the small
sample size and no external validation was performed, further independent multi-center studies are required to validate
our model using larger datasets from different sources as well as other biomarkers.

In this paper, we define a new class of generalized open sets in a generalized topological space
(????, ????), referred to as ℎ???????? − ???????????????? sets. We further explore topological concepts such as the ℎ???????? − ????????????????????????????????, ℎ???????? −
????????????????????????????, and ℎ???????? − ???????????????????????????????? based on these sets. Additionally, we introduce and examine several types of functions
including ℎ???????? − ????????????????????????????????????????, ℎ???????? −????????????????, ℎ???????? − ???????????????????????????????????? ????????????????????????????????????????, ℎ???????? − ???????????????????????? ???????????????????????????????????????? function,
and ℎ???????? − ℎ????????????????????????????????ℎ???????????????? . Various properties and remarks related to these functions are also discussed.
Furthermore, we introduce the concepts of ????????0ℎ????????, ????????1ℎ???????? and ????????2ℎ???????? − ????????????????????????, and explore their characterization
as well as the relationships among ????????0ℎ????????, ????????1ℎ???????? and ????????2ℎ???????? − ????????????????????????.

Employee turnover costs businesses as much as 150% of an employee's annual salary. Organizational culture
directly drives turnover and yet, current measurement systems ignore qualitative text feedback from the employee.
The study builds a hybrid information technology system that combines organizational culture and employee
turnover based on questionnaires data and text mining. A Total of 2000 employees were sent a closed-ended
(organizational culture assessment) & open-ended item questionnaire, administered online. Text preprocessing
included tokenization, stopword removal, and lemmatization. Implemented VADER (lexicon based) and fine-tuned
BERT for Sentiment Analysis. Through separate training, machine learning models (Random Forest, XGBoost,
Logistic Regression) trained on closed-ended data alone and combined with fastText-derived features (sentiment
scores and keyword indicators). Model performance was measured using accuracy, precision, recall, F1-score and
AUC- ROC. The hybrid XGBoost model (closed-ended + text features) had 87 per cent accuracy, 85 per cent
precision, 84 per cent recall, 0.84 F1-score and 0.89 AUC-ROC — a nine-percentage-point increase from closedquestions-only XGBoost (78% accuracy). The term "unfair" had a 4.2× higher turnover risk, "burnout" was
associated with a 3.8× higher risk, and "recognition" reflected 3.5× lower (protective) on turnover risk. Average
sentiment score of 2.8/5 with 34% classified as highly negative. The main advantage of marrying the questionnaires
with text mining is that predicts whether someone will leave the organization or not. The hybrid model of XGBoost
highlighted those interpretable risk multipliers that can be acted on by HR practitioners. Recommendation: Include
open-ended employee feedback in HR analytics systems. Domain-adaptive sentiment model and multilingual
capability are two subjects for future work.

Objectives: The objective of the study is to systematically analyse the stability of a variety of
nonlinear initial-boundary value problems (also referred to as Darbaugh problems) by utilizing fixed-point methods.
Methodology: A partial differential equation governing the problem may be represented in the form of an equivalent
fixed-point representation in a suitably chosen Banach algebra or Sobolev space. The Banach Contraction Principle,
Krasnoselskii’s Theorem, and Schaefer’s Theorem may then be used to demonstrate the stability of these types of
problems. Results: We present explicit, algebraically verifiable criteria for the stability of such problems.
Specifically, a Lipschitz-type condition, with respect to the norm of the associated linear version of the solution
operator, guarantees uniform Hyers–Ulam stability; a decomposition of the problem into contractive and compact
components yields Lagrange (invariant set) stability; a uniform bound on the input to output relationship for a
homotopy family guarantees bounded-input-bounded-output stability. Numerous numerical examples are provided
for both parabolic and hyperbolic Darbaugh problems that validate these criteria. Conclusions: The fixed-point
method provides an effective and unified approach for establishing the existence, uniqueness, and stability of
solutions, resulting in tight error bounds and invariant sets that are not typically obtained through classical energy
methods. Recommendations: Researchers are invited to apply this approach to stochastic versions of the Darbaugh
problem and higher order algebraic structures (e.g. partially ordered Banach spaces) to determine criteria for the
stability of positive solutions.