The article examines the issue of drug clustering. Initially, k classes are arbitrarily formed and the resulting training sample is pre-processed, then the similarities between the objects of each class are evaluated based on the proximity function and the criterion for evaluating the contribution of objects to the formation of their own class. Usually, it is in percentage and is the degree of mutual similarity of objects of each class. In the next steps of the algorithm, first, one object is taken from the first class, and by adding it to all k classes, the contribution of this object to this class is measured. The object will be left in the class which has the most contribution. This process is repeated several times in a row for all objects of the class. The process is stopped when the location of objects does not change and the degree of similarity exceeds the required percentage. As a result, the required clusters are formed.

Recent advances in the areas of artificial intelligence and machine learning have opened up new avenues for enhancing the practice of medical diagnosis. However, researchers face difficulties in accessing quality datasets because of the sensitive nature of real clinical data related to diabetes mellitus. The main objective of this research is to introduce an algorithm intended to generate a semi-synthetic training dataset aimed at improving classification accuracy for diabetes mellitus, particularly type 1 and type 2 diabetes. An algorithm to generate semi-synthetic diabetes data by statistically analyzing clinical attributes from real patient records. For improving the generation of synthetic samples without altering the properties of the original data, a similarity-based approach focusing on class-object relations was used. The approach used successfully generated synthetic data instances that preserved the inherent structure and distribution typical of real patient data. A similarity-based mechanism ensured the relevance of the created instances, while the study outlined a sequence of steps intended to improve the quality of synthetic datasets. The proposed algorithm creates artificial datasets for diabetes classification with patient data protection. This methodology led to the rise in intra-class similarity from 76.18% to 82.93%, which in turn enhanced the diagnostic accuracy of artificial intelligence-based models.

Using traditional methods such as ear tags, branding, and tattoos for cattle identification requires continuous human involvement and demands significant time and effort. Although radio-frequency identification (RFID) methods are widely used today, they also have certain disadvantages. The RFID devices used must be constantly installed, which can cause discomfort to the animals, and during their movements, the devices may become damaged or lost, leading to further issues.By using biometric features for cattle identification, these disadvantages are eliminated. In this method, animals are identified using their unique biometric features, such as iris patterns, skin textures, muzzle prints, and facial features. This article focuses specifically on identifying cattle based on their muzzle images. In this study, a total of 4923 images of 268 cattle were used. The architectures of eight different models were improved and selected for the training process, and the training was conducted. According to the results, the DonsoNot-121, WidoRosNot-50, and Inception V3 models achieved the highest accuracy rates, with 99.2%, 99.1%, and 99.1%, respectively. These results demonstrate the effectiveness of the proposed architecture.

Early detection of brain cancer is crucial for improving treatment outcomes and patient survival rates. This study proposes a set of algorithms for clustering, classification, and selection of informative features that can aid in the early diagnosis of brain tumours. In collaboration with medical experts, a dataset comprising 218 patient records and 82 clinical and symptomatic features was constructed. Through clustering analysis, the dataset was grouped into four diagnostic classes: (1) Anaplastic astrocytoma in the right frontal region, (2) Adenoma in the chiasmal-sellar region, (3) Glioblastoma in the right frontal lobe, and (4) Meningioma in the right frontal lobe. A feature selection algorithm was then applied to identify the most diagnostically relevant attributes. From the initial 82 features, 19 were determined to be strongly indicative of the disease classes. Further refinement using the proposed algorithm resulted in a subset of six highly informative features, which successfully differentiated the classes with a minimum object similarity threshold of 65 %. The approach demonstrates the potential of data-driven techniques in enhancing the accuracy and efficiency of brain cancer diagnostics, offering a scalable method that could be integrated into clinical decision-support systems.

Accurate classification of pollen grains in microscopic images remains challenging because of noise, structural variability, background complexity, weak texture, and intra-class similarity. To address these issues, this study proposes a hybrid framework that integrates contour-signal modeling, spectral–wavelet analysis, and deep learning for robust microscopic pollen image recognition. In the proposed approach, microscopic pollen images are first converted into contour-based point-signal representations, allowing object boundaries to be analyzed as structured one-dimensional signals. To improve signal quality under real imaging conditions, the framework incorporates Gaussian, median, and contour-aware filtering together with defect-point detection and correction. The processed contour signals are then analyzed using Fourier transform, continuous wavelet transform, and discrete wavelet transform to extract complementary global and local descriptors. These enriched representations are provided to a convolutional neural network for final classification. Experiments conducted on a seven-class microscopic pollen-image dataset demonstrate that the proposed method outperforms conventional computer-vision and baseline deep-learning approaches. The best-performing hybrid configuration achieved an error rate of 6.4%, while the overall classification accuracy reached 0.977 with an F1-score of 0.966, compared with 0.837 for a traditional computer-vision pipeline. These results confirm that combining contour-based signal processing with hierarchical deep feature learning provides an effective and noise-robust strategy for microscopic pollen image recognition. However, the present validation is limited to pollen images, and further experiments on broader microscopic object datasets are required to assess generalization to other micro-object categories such as nanoparticles, fibers, rods, and synthetic microstructures.