Goran Martinović

This paper examines how data preparation affects machine-learning classifiers for diabetes-risk prediction using the Pima Indians Diabetes Database. Three preprocessing methods are considered: imputing invalid zeros, handling outliers, and data scaling. Nine algorithms are evaluated on this dataset: linear/probabilistic baselines (Logistic Regression, Gaussian Naive Bayes), distance-based methods (KNN, Support Vector Machines), a single tree-based model (Decision Tree), and tree ensembles (Random Forest, Gradient Boosting, XGBClassifier, LightGBM). Median imputation of invalid zeros yields the largest and most consistent gains in accuracy and AUC. Outlier handling uses interquartile-range filtering, with Local Outlier Factor as an auxiliary indicator; effects are modest for accuracy and small, model-dependent for AUC. Scaling offers targeted benefits: for KNN, robust scaling can slightly alter performance and may reduce AUC relative to median-only imputation in this setup; SVM shows modest gains, while tree ensembles are comparatively insensitive overall. Ensembles achieve the highest performance and remain robust under minimal preparation, while simpler models benefit most from pipelines combining median imputation, careful outlier handling, and appropriate scaling. Hyperparameter tuning yields small to substantial gains—large for Decision Trees—while leaving ensemble rankings largely unchanged. Overall, results highlight the centrality of median imputation and the selective value of scaling for distance-based classifiers in diabetes-risk prediction.

Pneumonia remains a major global health concern, particularly among pediatric populations in low-resource settings where radiological expertise is limited. This study investigates the enhancement of deep convolutional neural networks (CNNs) for automated pneumonia diagnosis from chest X-ray images through the integration of a novel module combining Inception blocks, Mish activation, and Batch Normalization (IncMB). Four state-of-the-art transfer learning models—InceptionV3, InceptionResNetV2, MobileNetV2, and DenseNet201—were evaluated in their base form and with the proposed IncMB extension. Comparative analysis based on standardized classification metrics reveals consistent performance improvements across all models with the addition of the IncMB module. The most notable improvement was observed in InceptionResNetV2, where the IncMB-enhanced model achieved the highest accuracy of 0.9812, F1-score of 0.9761, precision of 0.9781, recall of 0.9742, and strong specificity of 0.9590. Other models also demonstrated similar trends, confirming that the IncMB module contributes to better generalization and discriminative capability. These enhancements were achieved while reducing the total number of parameters, indicating improved computational efficiency. In conclusion, the integration of IncMB significantly boosts the performance of CNN-based pneumonia classifiers, offering a promising direction for the development of lightweight, high-performing diagnostic tools suitable for real-world clinical application, particularly in underserved healthcare environments.

In recent years, Transformer-based models have dominated the field of long-term time series forecasting. However, the quadratic complexity of attention mechanisms makes both training and inference computationally expensive. The SOFTS model has emerged as an efficient alternative, replacing attention mechanisms with the STAR module to preserve linear complexity while achieving performance comparable to or better than competing approaches. The SOFTS model builds on the iTransformer architecture, which marked a significant advancement in long-term time series forecasting. Although neither iTransformer nor SOFTS incorporates positional embeddings, our analysis revealed a clear opportunity to improve forecasting accuracy by introducing them. However, the straightforward inclusion of positional embeddings leads to convergence and generalization issues. To address this, we propose a simple yet effective technique: during training, positional embeddings are randomly omitted in certain forward passes, which reduces instability and helps the model generalize better. We refer to this novel form of using positional embeddings as Learnable Stochastic Positional Embedding. Additionally, we incorporate multiple dropout layers to mitigate overfitting and improve accuracy. These modifications result in SOFTS++, a fast and accurate model that achieves the best performance on at least 10 out of 12 standard benchmark datasets. By maintaining linear complexity and requiring minimal computational resources, SOFTS++ stands out as a capable and resource-efficient method for multivariate long-term forecasting tasks.

Atypical and subtle facial expression patterns in individuals with autism spectrum disorder (ASD) pose a significant challenge for automated emotion recognition. This study evaluates and compares the performance of convolutional neural networks (CNNs) and transformer-based deep learning models for facial emotion recognition in this population. Using a labeled dataset of emotional facial images, we assessed eight models across four emotion categories: natural, anger, fear, and joy. Our results demonstrate that transformer models consistently outperformed CNNs in both overall and emotion-specific metrics. Notably, the Swin Transformer achieved the highest performance, with an accuracy of 0.8000 and an F1-score of 0.7889, significantly surpassing all CNN counterparts. While CNNs failed to detect the fear class, transformer models showed a measurable capability in identifying complex emotions such as anger and fear, suggesting an enhanced ability to capture subtle facial cues. Analysis of the confusion matrix further confirmed the transformers’ superior classification balance and generalization. Despite these promising results, the study has limitations, including class imbalance and its reliance solely on facial imagery. Future work should explore multimodal emotion recognition, model interpretability, and personalization for real-world applications. Research also demonstrates the potential of transformer architectures in advancing inclusive, emotion-aware AI systems tailored for autistic individuals.

Pediatric pneumonia remains a critical global health challenge requiring accurate and interpretable diagnostic solutions. Although deep learning has shown potential for pneumonia recognition on chest X-ray images, gaps persist in understanding model interpretability and feature learning during training. We evaluated four convolutional neural network (CNN) architectures, i.e., InceptionV3, InceptionResNetV2, DenseNet201, and MobileNetV2, using three approaches—standard convolution, multi-scale convolution, and strided convolution—all incorporating the Mish activation function. Among the tested models, InceptionResNetV2, with strided convolutions, demonstrated the best performance, achieving an accuracy of 0.9718. InceptionV3 also performed well using the same approach, with an accuracy of 0.9684. For DenseNet201 and MobileNetV2, the multi-scale convolution approach was more effective, with accuracies of 0.9676 and 0.9437, respectively. Gradient-weighted class activation mapping (Grad-CAM) visualizations provided critical insights, e.g., multi-scale convolutions identified diffuse viral pneumonia patterns across wider lung regions, while strided convolutions precisely highlighted localized bacterial consolidations, aligning with radiologists’ diagnostic priorities. These findings establish the following architectural guidelines: strided convolutions are suited to deep hierarchical CNNs, while multi-scale approaches optimize compact models. This research significantly advances the development of interpretable, high-performance diagnostic systems for pediatric pneumonia using chest X-rays, bridging the gap between computational innovation and clinical application.

Pediatric pneumonia remains a significant global health issue, particularly in low- and middle-income countries, where it contributes substantially to mortality in children under five. This study introduces a deep learning model for pediatric pneumonia diagnosis from chest X-rays that surpasses the performance of state-of-the-art methods reported in the recent literature. Using a DenseNet201 architecture with a Mish activation function and multi-scale convolutions, the model was trained on a dataset of 5856 chest X-ray images, achieving high performance: 0.9642 accuracy, 0.9580 precision, 0.9506 sensitivity, 0.9542 F1 score, and 0.9507 specificity. These results demonstrate a significant advancement in diagnostic precision and efficiency within this domain. By achieving the highest accuracy and F1 score compared to other recent work using the same dataset, our approach offers a tangible improvement for resource-constrained environments where access to specialists and sophisticated equipment is limited. While the need for high-quality datasets and adequate computational resources remains a general consideration for deep learning applications, our model’s demonstrably superior performance establishes a new benchmark and offers the delivery of more timely and precise diagnoses, with the potential to significantly enhance patient outcomes.

Pneumonia remains a significant cause of morbidity and mortality among pediatric patients worldwide. Accurate and timely diagnosis is crucial for effective treatment and improved patient outcomes. Traditionally, pneumonia diagnosis has relied on a combination of clinical evaluation and radiologists’ interpretation of chest X-rays. However, this process is time-consuming and prone to inconsistencies in diagnosis. The integration of advanced technologies such as Convolutional Neural Networks (CNNs) into medical diagnostics offers a potential to enhance the accuracy and efficiency. In this study, we conduct a comprehensive evaluation of various activation functions within CNNs for pediatric pneumonia classification using a dataset of 5856 chest X-ray images. The novel Mish activation function was compared with Swish and ReLU, demonstrating superior performance in terms of accuracy, precision, recall, and F1-score in all cases. Notably, InceptionResNetV2 combined with Mish activation function achieved the highest overall performance with an accuracy of 97.61%. Although the dataset used may not fully represent the diversity of real-world clinical cases, this research provides valuable insights into the influence of activation functions on CNN performance in medical image analysis, laying a foundation for future automated pneumonia diagnostic systems.

Due to the projected increase in food production by 70% in 2050, crops should be additionally protected from diseases and pests to ensure a sufficient food supply. Transfer deep learning approaches provide a more efficient solution than traditional methods, which are labor-intensive and struggle to effectively monitor large areas, leading to delayed disease detection. This study proposed a versatile module based on the Inception module, Mish activation function, and Batch normalization (IncMB) as a part of deep neural networks. A convolutional neural network (CNN) with transfer learning was used as the base for evaluated approaches for tomato disease detection: (1) CNNs, (2) CNNs with a support vector machine (SVM), and (3) CNNs with the proposed IncMB module. In the experiment, the public dataset PlantVillage was used, containing images of six different tomato leaf diseases. The best results were achieved by the pre-trained InceptionV3 network, which contains an IncMB module with an accuracy of 97.78%. In three out of four cases, the highest accuracy was achieved by networks containing the proposed IncMB module in comparison to evaluated CNNs. The proposed IncMB module represented an improvement in the early detection of plant diseases, providing a basis for timely leaf disease detection.