Browsing by Author "Unlu, Ramazan"

Now showing 1 - 4 of 4

Citation - WoS: 1
Citation - Scopus: 1
Prediction of Biomechanical Properties of Ex Vivo Human Femoral Cortical Bone Using Raman Spectroscopy and Machine Learning Algorithms
(Elsevier, 2025) Unal, Mustafa; Unlu, Ramazan; Uppuganti, Sasidhar; Nyman, Jeffry S.
This study applied Raman spectroscopy (RS) to ex vivo human cadaveric femoral mid-diaphysis cortical bone specimens (n = 118 donors; age range 21-101 years) to predict fracture toughness properties via machine learning (ML) models. Spectral features, together with demographic variables (age, sex) and structural parameters (cortical porosity, volumetric bone mineral density), were fed into support vector regression (SVR), extreme tree regression (ETR), extreme gradient boosting (XGB), and ensemble models to predict fracture-toughness metrics such as crack-initiation toughness (Kinit) and energy-to-fracture (J-integral). Feature selection was based on Raman-derived mineral and organic matrix parameters, such as nu 1Phosphate (PO4)/CH2-wag, nu 1PO4/ Amide I, and others, to capture the complex composition of bone. Our results indicate that ensemble models consistently outperformed individual models, with the best performance for crack initiation toughness (Kinit) prediction being achieved using the ensemble approach. This yielded a coefficient of determination (R2) of 0.623, root-mean squared error (RMSE) of 1.320, mean absolute error (MAE) of 1.015, and mean percentage absolute error (MAPE) of 0.134. For prediction of the overall energy to propagate a crack (J-integral), the XGB model achieved an R2 of 0.737, RMSE of 2.634, MAE of 2.283, and MAPE of 0.240. This study highlights the importance of incorporating mineral quality properties (MP) and organic matrix properties (OMP) for enhanced prediction accuracy. This work represents the first-ever study combining Raman spectroscopy with other clinical and structural features to predict fracture toughness of human cortical bone, demonstrating the potential of artificial intelligence (AI) and ML in advancing bone research. Future studies could focus on larger datasets and more advanced modeling techniques to further improve predictive capabilities.
Citation - WoS: 1
Citation - Scopus: 1
Strategic Investment in BIST100: A Machine Learning Approach Using Symbolic Aggregate Approximation Clustering
(Univ Cincinnati industrial Engineering, 2025) Nalici, Mehmet Eren; Soylemez, Ismet; Unlu, Ramazan
This study employs the Symbolic Aggregate Approximation (SAX) clustering method to enhance investor decision-making on the Borsa Istanbul (BIST100) by identifying companies exhibiting analogous stock movements. The data from 81 BIST100 companies over a three-year period has been analyzed, with a focus on risk minimization and strategic investment. The SAX method, integrated with a dendrogram, categorizes stocks into sector-based and non-sector-based clusters, providing insights for portfolio optimization. The results demonstrate the effectiveness of the method in identifying relevant stock patterns across sectors, aiding in more informed investment decisions. This approach highlights the need for considering multiple factors in investment strategies, offering a new perspective on stock market analysis with advanced clustering techniques.
Fluctuations in the European Housing Market: Forecasting the House Price Index Change with Time-Series Models
(Gazi Univ, 2025) Soylemez, Ismet; Nalici, Mehmet Eren; Unlu, Ramazan
This study presents a comparative analysis of a time series models for forecasting changes in the Housing Price Index (HPI) in 27 European countries. Accurate HPI forecasting is essential for the development of effective policies and investment strategies. The study uses quarterly data from Q4 2013 to Q3 2024. Methodologically, the stationarity of the data is tested using the Dickey-Fuller test and differencing is applied to non-stationary series. The ARIMA, Holt Linear Trend, Additive Damped Trend and Exponential Smoothing models are evaluated based on the lowest mean squared error (MSE) value for each country. The findings confirmed the heterogeneous structure of the European housing market, showing that no single model is suitable for all countries. The ARIMA model provided the most accurate results for nine countries, while the Holt Linear Trend and Additive Damped Trend models performed best in seven countries each. Forecasts for the period 2025-2026 are generated based on these results. This study highlights the importance of adopting country-specific and adaptable forecasting approaches to accommodate the varying dynamics of European housing markets.
High-Accuracy Identification of Durian Leaf Diseases: A Convolutional Neural Network Approach Validated with K-Fold Cross-Validation and Bayesian Optimization
(Springer, 2025) Soylemez, Ismet; Nalici, Mehmet Eren; Unlu, Ramazan
To address the economic losses caused by plant diseases in durian farming, this study presents an optimized deep learning model that diagnoses diseases from leaf images with high accuracy. The model's performance is maximized through Bayesian optimization and hyperparameter tuning, while its reliability is maximized through layered five-fold cross-validation. Training the convolutional neural network model on 2595 leaf images displaying six different states (five diseased and one healthy) resulted in an average test accuracy of 91.98%. This high, consistent success rate demonstrates the model's generalizability to different datasets without overfitting. While the 'Healthy' and 'Algal' classes were successfully detected with high F1-scores, there are difficulties distinguishing between the 'Blight' and 'Colletotrichum' classes due to visual similarities. This study establishes a new reference point for durian disease classification and makes a significant contribution to the development of reliable artificial intelligence-based diagnostic tools for precision agriculture.