MEMustafa Etcil

Mustafa Etcil

*This is an algorithmically generated author record based on publicly available information.

Current Organizations / Institutes

The organizations/institutes from the scholar's latest article


Mustafa Etcil has a total of 12 co-authors. Here are the top co-authors by number of shared publications:


Latest Publications

  • Open Access

    AN EXPLAINABLE MACHINE LEARNING APPROACH USING FUNDAMENTAL RATIOS FOR STOCK RETURN PREDICTION AND PORTFOLIO REBALANCING

    by Mustafa Etcil, Burak Kolukisa, Burcu Güngör

    This research develops an explainable ensemble learning framework to forecast stocks’ market-relative performance and guide portfolio rebalancing in Borsa Istanbul (BIST). The analysis covers quarterly data from 20 firms between 2009-Q1 and 2025-Q1 and employs Random Forest, XGBoost, LightGBM, and CatBoost trained on thirteen financial ratios. Hyperparameters are selected using grid search with time-series cross-validation, and performance is evaluated out-of-sample. Portfolios are rebalanced quarterly using predicted returns and benchmarked against an equally weighted portfolio and the BIST 100 Index. Results show that model-based portfolios generally outperform benchmarks in total return and risk-adjusted performance, with stronger statistically supported results in broader Top-k configurations, particularly CatBoost Top-7. SHAP identifies profitability and valuation indicators, especially return on assets, return on equity, and key valuation ratios, as the main drivers of future stock performance. These findings support explainable ensemble learning for fundamental-based portfolio management in emerging markets.

  • Evaluation of Adversarial Robustness in Deep Learning-Based Colorectal Histopathology Classification

    by Tuğba Karataş, Mustafa Etcil, Burak Kolukisa

    Background/Aim: Deep learning models can achieve high accuracy in colorectal histopathology classification, but their performance may decrease under adversarial perturbations. This study aimed to evaluate the adversarial robustness of a 2D-CNN across two colorectal histopathology datasets under FGSM, PGD, and BIM attacks and to compare adversarial training with convolutional autoencoder (CAE)-based input purification as defense strategies.Methods: Experiments were conducted on the Kather Texture 2016 and CRC-VAL-HE-7K datasets, containing 5,000 and 7,180 image patches from eight and nine colorectal tissue classes, respectively. A 2D-CNN was trained using five-fold stratified cross-validation and evaluated on held-out test sets. White-box FGSM, PGD, and BIM attacks were applied across multiple perturbation levels. Adversarial training was examined using three clean-to-adversarial training compositions. For input purification, shallow, deep, and mixed-ε deep CAE models were evaluated. Performance was assessed using accuracy, macro precision, macro recall, and macro F1-score. Results: The 2D-CNN achieved 87.70% and 89.35% clean test accuracy on Kather Texture 2016 and CRC-VAL-HE-7K, respectively, but performance declined markedly under adversarial attacks, particularly on CRC-VAL-HE-7K. Adversarial training substantially improved robustness at ε = 0.01, although the best training composition varied across datasets and attacks. Mixed-ε deep CAE generally provided the strongest purification performance at higher perturbation levels. Conclusion: Despite its clean-image classification performance, the 2D-CNN remained vulnerable to adversarial perturbations. Both defense strategies generally improved performance under the evaluated adversarial conditions: adversarial training substantially improved performance at ε = 0.01, whereas mixed-ε CAE purification generally showed stronger relative performance at higher perturbation levels.

  • Investigating strain rate effects on damage mechanisms in hybrid laminated composites using acoustic emission

    by Abdulkadir Gulsen, Abdulkadir Gulsen, Burak Kolukisa, Burak Kolukisa, Mustafa Etcil, Mustafa Etcil, Umut Caliskan, Umut Caliskan, Hafız Muhammad Numan Zafar, Hafız Muhammad Numan Zafar, et al.


  • Open Access

    Breast Cancer Detection Using a New Parallel Hybrid Logistic Regression Model Trained by Particle Swarm Optimization and Clonal Selection Algorithms

    by Mustafa Etcil, Bilge Kagan Dedeturk, Burak Kolukisa, Burcu Bakir‐Gungor, Vehbi Cagri Gungor

    Breast cancer is one of the most widespread kinds of cancer, especially in women, and it has a high mortality rate. With the help of technology, it is possible to develop a computer‐aided method for the diagnosis of breast cancer, which is crucial for effective treatment. Recent breast cancer diagnosis studies utilizing numerous machine learning models were efficient and innovative. However, it has been observed that they may have problems such as long training times and low accuracy rates. To this end, in this study, we present a new classifier that utilizes a hybrid of the clonal selection algorithm (CSA) and the particle swarm optimization (PSO) algorithm for the training of the logistic regression (LR) model, which is named CSA‐PSO‐LR. The proposed method is evaluated using two publicly accessible breast cancer datasets, that is, the Wisconsin Diagnostic Breast Cancer (WDBC) database and the Wisconsin Breast Cancer Database (WBCD), with 10‐fold cross‐validation and Bayesian hyperparameter optimization techniques. Additionally, a CPU parallelization method is applied, which substantially shortens the training time of the model. The efficacy of the CSA‐PSO‐LR classifier is compared with state‐of‐the‐art machine learning algorithms and related studies in the literature. Performance analysis indicates that the proposed method achieves 98.75% accuracy and 98.27% F1‐score on the WDBC dataset, and 97.94% accuracy and 97.35% F1‐score on the WBCD dataset. These results demonstrate the potential of the proposed method as an effective approach for improving breast cancer diagnosis.


  • Evaluating the Impact of Sentiment Analysis on Deep Reinforcement Learning-Based Trading Strategies

    by Mustafa Etcil, Burak Kolukisa, Burcu Bakir-Gungor

    Portfolio optimization is a form of investment management that aims to maximize returns while minimizing risks. However, the inherent complexity and unpredictability of financial markets pose a challenge. Recent advancements in machine learning, particularly in deep reinforcement learning (DRL), offer promising solutions by enabling dynamic and adaptive trading strategies. This paper presents a comprehensive evaluation of three actor-critic-based DRL algorithms-Advantage Actor-Critic (A2C), Deep Deterministic Policy Gradient (DDPG), and Proximal Policy Optimization (PPO)-applied to portfolio optimization. These strategies were implemented in both sentiment-aware and non-sentiment-aware versions, allowing for a direct comparison of their performance. The sentiment-aware models incorporated sentiment analysis using FinBERT and knowledge graphs to measure market sentiment from financial news, while the non-sentiment-aware models relied solely on stock prices and technical indicators. Our comparative study demonstrates that incorporating sentiment analysis resulted in consistently superior risk-adjusted returns and portfolio resilience during market fluctuations compared to non-sentiment-aware strategies.