https://isee.ui.ac.ir/ Computational Intelligence in Electrical Engineering en daily 1 Sun, 23 Aug 2026 00:00:00 +0330 Sun, 23 Aug 2026 00:00:00 +0330 https://isee.ui.ac.ir/article_30787.html definitive cure, early diagnosis can significantly slow the disease's progression. In this research, a framework based on transfer learning is presented for diagnosing different stages of Alzheimer's disease from brain MRI images. To reduce the effect of data imbalance, the SMOTE technique was used to balance the classes, and then deep features were extracted using the pre-trained InceptionV3 network. Subsequently, final classification was performed using the DEMNET architecture, which was selected due to its lower number of parameters and strong generalization capability for small medical datasets. Experimental results on the Kaggle dataset showed that the proposed model achieved an accuracy of 98.52%, precision of 98.31%, recall of 98.47%, and an F1-score of 98.39%. Furthermore, the area under the ROC curve (AUC) value was 99%, indicating the model's high discriminative ability in identifying different stages of the disease. A comparison of the results with those obtained without using SMOTE also indicates a significant improvement in all evaluation metrics. The findings suggest that the combination of transfer learning, data balancing, and the lightweight DEMNET architecture can serve as an efficient approach for the automatic diagnosis of Alzheimer's from MRI images. This approach holds promise for application in clinical decision support systems. https://isee.ui.ac.ir/article_30880.html The operation of distribution networks, due to uncertainties arising from the presence of renewable resources and responsive loads, requires the provision of flexibility as a novel approach in smart distribution networks. In this paper, flexibility provision is formulated as a bi-level optimization problem. At the upper level, the secure network operation model is presented. This level incorporates incentive pricing tariffs for responsive loads while covering chance constraints that guarantee network security indices. The chance constraints used at this level are modeled to rely on actual data rather than pre-defined probability distributions such as the normal distribution. In the process of converting chance constraints into deterministic constraints, machine learning methods such as linear regression and neural networks are employed. The lower level of the problem addresses the optimization of responsive loads' profit from flexibility provision. To transform the aforementioned bi-level problem into a solvable single-level problem, the lower-level optimization problem is replaced with its global optimality conditions and modeled as an equivalent constraint optimization problem. A 33-bus network with renewable resources is used to evaluate the proposed structure. The results indicate that the reliability of the neural network method in providing flexibility using dynamic tariffs is higher than other methods.