Biomedical images, whether acquired by techniques such as magnetic resonance imaging (MRI), computed tomography (CT), ultrasound, X-ray, or other methods, are commonly obtained and permanently stored for diagnostic purposes. Therefore, leveraging this large number of images has become essential for the development of intelligent medical diagnostic systems. In this work, we propose a new biomedical image recognition in two stages: the first stage is to introduce a new image feature extraction technique using quaternion Legendre orthogonal moments (QLOMs) to extract features from biomedical images. The second stage is to use radial basis function (RBF) neural networks for image classification to know the type of disease. To evaluate our computer-aided medical diagnosis system, we present a series of experiments were conducted. Based on the results of a comparative study with recent approaches, we conclude that our method is very promising for the detection and recognition of dangerous diseases.

The waste-to-energy process relies on the recovery and use of energy produced by the combustion of waste, which allows for waste treatment while ensuring energy production. To improve the prediction accuracy of energy recovered from waste combustion, a new prediction model based on orthogonal Laguerre polynomials is proposed in this work. Exploiting the orthogonality of Laguerre polynomials reduces approximation errors and facilitates parameter estimation through the calculation of orthogonal moments. To validate the proposed method, the propose prediction model was trained on a dataset published online and then used to predict the higher heating value (HHV) of municipal solid waste (MSW), in order to recover energy from this waste. A comparative study was conducted with predictive methods used in the field of energy recovery from waste. Evaluation metrics such as the coefficient of determination (R²), the root mean square error (RMSE), and the mean absolute percentage error $\left(MAPE\right)$ were used in this comparative study. The results obtained show that the proposed model achieves its best performance for a polynomial of degree $n=5,$ with an optimal value of $\alpha =0.9$ determined by the Artificial Bee Colony (ABC) algorithm. In this configuration, the model obtains a minimum RMSE of 1.09, compared to 5.12 for the LSTM-MLP model, which is the highest-performing reference method among those compared. These results demonstrate the ability of the proposed model to predict the higher calorific value HHV of municipal solid waste with significantly greater accuracy.

State-of-the-art Sentiment Analysis (SA) models that achieve high performance typically require substantial training datasets. To address this limitation, this paper introduces novel and efficient K-Nearest Neighbors (KNN) variants for SA, particularly for scenarios involving limited, sparse, and high-dimensional data. Firstly, we propose the Orthogonal Projection-based Iterative KNN classifier (OPIKNN), where the classification of unseen samples is performed based on their proximity to the pseudo-nearest neighbors of the respective classes. In the proposed classifier, the extreme-nearest neighbors of the unseen samples are identified iteratively to determine the pseudo-nearest neighbor using the orthogonal projection, which helps capture the class distribution more effectively. Secondly, building upon OPIKNN, we develop the Fuzzy OPIKNN (FOPIKNN) classifier to address the problem of data uncertainty and reduce the effects of noise in limited datasets. FOPIKNN uses a refinement step that shifts pseudo-nearest neighbors toward class centroids based on their fuzzy membership degree to these classes to provide a robust class representation. Extensive experimental results on six real-world SA datasets from Kaggle and Amazon demonstrate the effectiveness and reliability of the proposed approaches in comparison with six state-of-the-art KNN-based variants and four benchmark Machine Learning (ML) classifiers. In particular, the classification accuracy achieved by OPIKNN and FOPIKNN improves, on average, by 5.8% and 7.3%, respectively, relative to the standard KNN classifier. Even more notably, the macro \(F_1\)-score increases by 8.5% and 9.8%, respectively, demonstrating consistent performance gains across both balanced and imbalanced data distributions.

This study presents a hybrid Non-Uniform Rational B-Spline (NURBS) methodology for the geometric reconstruction of topology-optimized structural components. NURBS are employed exclusively as a post-processing tool; all structural analyses are performed using standard finite elements (SOLID187 elements, ANSYS Mechanical R19.2), and isogeometric analysis (IGA) is not used. The methodology is validated on an Al 6061-T6 electric scooter neck under a 600 N static load. Two SIMP optimization iterations followed by a hybrid NURBS reconstruction reduce the component mass from 1.247 kg to 0.531 kg, achieving a 57.4% mass reduction. Finite element re-validation of the reconstructed geometry yields a maximum von Mises stress of 126.45 MPa (safety factor, SF = 2.18, exceeding the 2.0 requirement), a maximum deflection of 2.31 mm, and a first natural frequency of 127 Hz. Mesh convergence between the 2.5 mm and 1.25 mm refinements is Δ = 0.90%. Relative to the direct SIMP output (201 MPa), NURBS reconstruction reduces the peak stress by 37%, demonstrating that geometric post-processing is not a neutral step but a critical determinant of structural performance. Both fully automated STL reconstruction and edge-based NURBS reconstruction failed for this geometry class due to non-manifold topology and patch discontinuities, respectively. The proposed hybrid region-decomposition approach is the only method that has produced a watertight, FEA-compatible CAD model.