Early detection of diabetes is critical in preventing disease progression and improving patient outcomes. This work combines Explainable Artificial Intelligence (XAI) and machine learning to enhance understanding and prediction of diabetes using the Pima Indian Diabetes Dataset. The machine learning models used in this study are Random Forest, Logistic regression and Gradient Boosting, which resulted in the best accuracy of 93.2%. Some of the pre-processing steps taken were handling of missing data, normalization, feature scaling and Synthetic Minority Over-sampling Technique (SMOTE) for handling class imbalance. Use of SHAP and LIME XAI methods has proven that glucose, BMI and insulin are the most crucial features when it comes to prediction. These techniques further enhance the trust of the clinicians and stakeholders by improving the understanding of how the features contribute to individual predictions, which enhances the model prediction as a whole. The findings prove that there is indeed a marked improvement in the understanding of the machine learning models and their predictions with no compromise on performance. This study highlights the benefits of using XAI in machine learning so that there is accuracy and ease of interpretation with immense power within the developed models.
An optimal investment portfolio is one of the main focuses in the financial world to minimize risk while maximizing returns. However, the challenge that arises is how to choose the right asset allocation amidst dynamic market uncertainty. This study aims to optimize portfolios based on Markowitz modern portfolio theory (MPT) by using the differential evolution (DE) algorithm as an optimization technique. The data used includes stocks, bonds, and other financial instruments taken from trusted data sources, such as Bloomberg and Yahoo finance, with an observation period of the last five years. The results show that this approach succeeds in finding optimal portfolios with the right asset weights, higher expected returns, and minimized risks compared to conventional approaches. The implication of this research is that the DE algorithm can be effectively used to address portfolio optimization problems in complex and volatile market environments, offering a more adaptive solution for investors to maximize their returns.
Tuberculosis (TB) affects the health of many individuals and is still a prime worldwide health concern despite having so many advanced treatments, as it still lacks technical advancement in its treatment and diagnosis. Accuracy in identification and early detection is essential to reduce the spread and improve treatment outcomes. Traditional methods of diagnosis, such as sputum microscopy and culture, are labor-dependent and subject to human mistakes as it is done by lab technicians. Recent improvements in deep learning have demonstrated significant potential for enhancing and automating diagnostic accuracy. Our research proposes a deep learning based technique that detects TB from chest X-rays after image processing techniques like augmentation. After training on big data, our model pulls off an astonishing accuracy of 97.42% and a loss of 7.17%, outperforming traditional methods. The model uses convolutional neural network (CNN) as a base and transfer learning method, like DenseNet-121, and explainable artificial intelligence (XAI) technique, like Grad-CAM, to recognize TB related patterns effectively and with low false positives. This approach has the ability to revolutionize the diagnosis of TB and offer more dependable, scalable, and timely solutions to healthcare systems worldwide.
Large language models (LLMs) have transformed natural language processing (NLP), demonstrating exceptional proficiency in tasks such as text generation, translation, and summarization. However, LLMs are prone to generating biased, inaccurate, or contextually irrelevant outputs, posing significant risks in high-stakes domains such as healthcare, legal reasoning, and engineering. This paper systematically investigates the role of prompt engineering as a solution to these challenges. By strategically designing inputs, prompt engineering enhances LLM performance, yielding more accurate, contextually relevant, and ethically aligned outputs. Advanced techniques, including chain-of-thought (CoT) prompting and retrieval augmented generation (RAG), are examined for their ability to improve reasoning capabilities, reduce errors, and mitigate bias. CoT prompting facilitates structured, stepwise reasoning, while RAG incorporates real-time data, ensuring output accuracy in rapidly evolving fields. In addition, we present a novel comparative perspective on these techniques, highlighting their distinct strengths and limitations across specialized applications such as healthcare diagnostics and scientific data extraction. The findings demonstrate that sophisticated prompt engineering significantly elevates the reliability and precision of LLM outputs, while addressing critical ethical concerns such as data privacy, bias, and hallucination. These insights underscore the necessity of advanced prompt design in optimizing LLMs for high-impact applications, ensuring both performance and ethical integrity.
Smart light controllers revolutionize energy efficiency and convenience by controlling lights with precision. The main problem with such control structures is that the existing lighting systems are outdated and inefficient in places such as libraries, leading to unnecessary energy consumption. In this study, we propose the implementation of an intelligent light-emitting diode (LED) lighting control system that optimizes energy in libraries. The system utilizes passive infrared (PIR) sensors and a counter mechanism. When an aisle is occupied, the LEDs operate in highlight mode; however, they switch to medium-light mode when no presence is detected for a predetermined period. For this process, we propose a new control system with the random forest prediction algorithm for displacement patterns. We also designed a mathematical algorithm with three matrices derived from previous work on lighting. The algorithmic process is applied to ensure that the proposed system has enhanced maintainability, the simulation methodology generated scenarios based on occupancy data collected from real library environments, MATLAB was used to create dynamic simulations that reflected these user behaviors, with the power consumption of each LED calculated over an eight-hour operating period, the proposed system achieves potential savings between 23% to 39%.
Despite the prevalence of this disease, the existing method for obtaining an exact breast cancer diagnosis would need a lot of time and labor. It needs a qualified pathologist to manually process and review histopathological images to distinguish the characteristics that characterize different cancer severity levels. Building a model for automatically detecting, segmenting, and classifying breast lesions using histopathological images seems to be the goal of this work. Various deep learning methods have been used in computational pathology for the diagnosis of cancer. Improved faster recurrent convolutional neural network (IMFRCNN) is a supervised learning system with proposed for recognizing small items like mitotic and non mitotic nuclei. To protect small items from vanishing in the deep layers, this system uses expanded layers in the spine. To close image and the things gap size includes, this approach uses expanded layers. The region proposal network has been created for precise tiny object identification. Researchers examined time for training and testing time for various techniques for identifying objects. The total accuracy of benign/malignant categorization in proposed system reaches 96.5%. The proposed technique offers a thorough and non-invasive method for identifying and categorizes an area of abnormal breast tissue.
The current research will create a robust and successful deep learning (DL) system to recognize and classify papaya leaf diseases. The traditional disease detection techniques are both time-consuming and unreliable, and extensively rely on expert knowledge, therefore limiting them in terms of scalability in agricultural practice. To tackle this issue, the convolutional neural network (CNN)-based method is suggested and tested on the BDPapayaLeaf that includes 2,159 images of papaya leaf with four disease categories and healthy papaya leaves, i.e., anthracnose, bacterial spot, leaf curl (reversal), and ring spot. The data was split into training 80%, validation 10%, and testing 10% data. Pictures were downscaled to 224×224 and normalized before training. Six trained CNN structures VGG16, VGG19, InceptionV3, DenseNet121, MobileNetV2, and ResNet50 were examined. The top model in terms of classification accuracy, according to them, was InceptionV3 with 89% in terms of classification accuracy, showing a high level of performance on true positive and false negative. The findings indicate that DL is an effective and precise method of automated detection of papaya leaf disease and is useful in improving precision and reliability in agricultural diagnostics.
Cybercrime gained traction in the late 20th century. The capabilities of cyber-attackers have improved dramatically. One of the biggest challenges facing cybersecurity developers is safeguarding consumers' security and privacy. Interest in using AI approaches in cybersecurity has grown significantly because of the incredible proficiency these techniques have demonstrated across all domains. Even while machine learning algorithms are very effective at identifying malicious activity, there are still certain issues that lower performance accuracy. This paper has the novelty of deploying the Artificial Bee Colony (ABC) meta-heuristic algorithm with the K-Nearest Neighbors (KNN) classifier to detect cyber-attacks. It proposes a variant approach called KNN+Bee that detects attacks efficiently, achieving 99.86% overall accuracy. The NSL-KDD dataset of cyberattacks has been leveraged in the training and testing phases. The proposed approach has been contrasted with the most popular machine learning. According to experimental findings, the suggested model delves deeper into the identification of cyberattacks. It achieves unprecedented performance, outperforming other models in terms of precision, Recall, F-score and MCC. Furthermore, popular deep learning models have been implemented and examined on the same dataset. Results prove that GRU is the most accurate, reaching 99.71%.
The rapid advancements in fetal ultrasonography have significantly enhanced prenatal diagnosis in recent years. Deep learning (DL) architectures have further streamlined the process of organ detection, improved diagnostic accuracy, and reduced observer dependency. This study proposes a computer-aided DL approach for fetal organ segmentation using the you only look once (YOLO) algorithm, a state-of-the-art method for object detection and image segmentation. This study identified and classified 15 fetal organs, including the umbilical vein, stomach, abdomen, brain (trans-cerebellum, trans-thalamic, and trans-ventricular regions), femur, head, thorax (chest cavity), heart (circumference, left atrium, left ventricle, right atrium, right ventricle), and aorta. We compared the performance of YOLOv7, YOLOv8, YOLOv9, and YOLOv11 architectures. The results showed that YOLOv9 outperformed YOLOv7, YOLOv8, and YOLOv11 achieving mAP50 and mAP95 scores of 91.90% and 94.50%, respectively. This performance surpasses previous studies that focused on classifying only a limited number of fetal organs.
Rider fatigue poses a critical safety challenge in two-wheeled vehicle operation due to limited physical protection, increased balance demands, and prolonged exposure to environmental stressors. Effective real-time fatigue detection is essential to mitigate accident risks, particularly in high-traffic regions such as Indonesia. This study presents a comparative analysis of black-box and glass-box artificial intelligence (AI) models for real-time detection of rider fatigue, evaluated through a human factor’s lens emphasizing interpretability, intrusiveness, and cognitive compatibility. Multimodal data comprising physiological signals, behavioral indicators, and environmental context were collected using wearable sensors and rider telemetry to train and assess the models. Experimental results reveal that black-box models, including convolutional neural network (CNN) + long short-term memory (LSTM), random forest (RF), and support vector machine (SVM), achieve superior predictive accuracy (94.3%, 91.5%, and 88.2%, respectively) but lack inherent transparency. Conversely, glass-box models such as decision tree (DT) and logistic regression (LR) offer greater interpretability, a critical factor in safety-sensitive applications, though with reduced accuracy (approximately 83–85%). These findings underscore the trade-off between predictive performance and explainability, highlighting the need to tailor model choice to specific operational requirements. This research advances the design of intelligent, human-centered rider support systems that balance accuracy, transparency, and user trust, fostering safer two-wheeled transportation.
In this study, we created and applied novel concepts for hardware-based image identification and categorization. For artificial intelligence (AI) and image recognition applications, this includes putting algorithms for recognizing colors, textures, and shapes into practice. Our contribution uses an embedded device with a camera and a microcomputer (Raspberry-Pi4 type) to replace the optical assessment of Petri dishes. Our object recognition system processes images efficiently by using a state-of-the-art kernel function and a new neighborhood architecture. Using the well-known convolutional neural network (CNN) architecture, YOLOv8, as a pre-trained model, we evaluated the proposed CNN-based method for object recognition in a number of demanding scenarios. Several Petri plates, uncontrolled settings, and different backgrounds and illumination were used to evaluate the technology. Our dynamic mode integrates a CNN network with an attention mask to highlight the traits of bacteria and fungi, ensuring robust recognition. We implemented our algorithm on a Raspberry Pi 400, connected to a CMOS 3.0 camera sensor and a human-machine interface (HMI) for instant display of results.
This paper presents a comprehensive framework for structured data collection and deep learning (DL)-based translation of retinal optical coherence tomography (OCT) images into diagnostic text. The suggested approach guarantees high-quality OCT data for model training through the use of sophisticated image processing methods like edge detection, noise suppression, and contrast improvement. The study utilizes 84,484 retinal images from the OCT dataset available on Kaggle. The research utilizes various preprocessing techniques, such as median and Gaussian filtering, along with data augmentation strategies like translation, rotation, and scaling, to mitigate class imbalances and improve model performance. The system automatically identifies and categorizes retinal diseases such as drusen, diabetic macular edema (DME), and choroidal neovascularization (CNV) by integrating feature extraction and selection with DL techniques. The research highlights the importance of effective data handling and model scalability to address the increasing need for automated diagnostic tools in ophthalmology. This framework aims to support ophthalmologists in managing the increasing incidence of diabetic retinopathy (DR) and other retinal conditions by enhancing the efficiency of retinal image analysis, thereby improving patient results through early detection and treatment.
Gait disorders in adults aged 50 years and above are a common concern and are often linked to reduced mobility, a higher risk of falls, and a lower quality of life. This study presents a deep learning-based approach to detect gait disorders using vertical ground reaction force (vGRF) signals. The data were collected from older adults, including individuals with Parkinson’s disease (PD) and healthy controls, using force-sensitive resistor sensors. The raw signals were first processed using band-pass filtering and wavelet denoising to remove noise and unwanted variations. After that, the signals were converted into time–frequency representations using the continuous wavelet transform (CWT). These representations were then used as input to a convolutional neural network (CNN) for classification. The model achieved a validation accuracy of 93.48%, with precision, recall, and F1-score all above 92% for both groups. The results show that combining CWT with CNN provides a reliable and efficient way to detect gait disorders. This approach can support clinical evaluation by offering a practical and scalable method for analyzing gait patterns in older adults.
Diabetic retinopathy (DR) is a complication caused by poorly managed diabetes that affects the eyes. According to the World Health Organization (WHO), 422 million people worldwide have suffered from DR in the past ten years. Manual detection using retinal fundus images is time-consuming and requires experienced ophthalmologists. This study proposes a deep learning method using the pre-trained model EfficientNet-B7 to identify this disease automatically. Five levels of DR will be classified: no-DR, mild-DR, moderate-DR, severe-DR, and proliferative-DR. The model was trained using "APTOS 2019 blindness detection" dataset, and image augmentation was performed. Image segmentation techniques such as contrast limited adaptive histogram equalization (CLAHE) and real enhanced super resolution generative adversarial network (Real-ESRGAN) were applied during preprocessing to improve the model's accuracy significantly. The implementation of CLAHE resulted in the validation accuracy improvement from 76.6% to 83.4% compared to no segmentation, while the combination of Real-ESRGAN and CLAHE increased the accuracy to 93.7%. Future research can explore the combination of CLAHE with other image processing techniques apart from the Real-ESRGAN model.
Apple bruise detection plays a crucial role in post-harvest quality control; however, conventional manual inspection remains labor-intensive, subjective, and unsuitable for large-scale industrial deployment. This study proposes an automated classification framework for identifying bruised regions in apples using hyperspectral imaging combined with deep learning and adaptive optimization techniques. The proposed model integrates a long short-term memory (LSTM) network optimized using an adaptive sand cat swarm optimization (ASCSO) algorithm, along with a ResNet-50 feature extraction backbone. The adaptive behavior embedded within ASCSO dynamically adjusts the optimization parameters to enhance convergence and prevent premature stagnation during LSTM hyperparameter tuning. Hyperspectral images were processed to extract relevant spectral–spatial features, which were subsequently fed into the optimized classifier. Experimental evaluations demonstrate that the proposed hybrid model significantly outperforms conventional and baseline deep learning approaches, achieving a classification accuracy of 98.0% while maintaining robustness across varying bruise patterns and intensity levels. The results highlight the effectiveness of combining hyperspectral imaging with adaptive deep learning optimization for high-precision fruit quality assessment. This research contributes a reliable, scalable solution for automated bruise detection and quality grading in the fruit supply chain, offering strong potential to reduce post-harvest losses and improve operational efficiency in the agro-food industry.
Stroke is a leading cause of death and disability worldwide, making early risk prediction critical for prevention. Machine learning methods such as random forest (RF) have shown strong predictive performance, but accuracy can be further improved through effective feature selection. This research proposes an integrated model that combines binary particle swarm optimization (BPSO) for feature selection with RF for stroke risk classification. Experiments were conducted on two public datasets: the stroke prediction dataset (SPD) and the brain stroke dataset (BSD). Data preprocessing included handling missing values, normalization, and the synthetic minority oversampling technique (SMOTE) to mitigate the minority and majority classes. BPSO was employed to select the most informative features, followed by RF for classification. The BPSO-RF model delivered superior accuracies of 96.13% on the SPD and 96.07% on the BSD, outperforming competing classifiers and feature selection techniques. Important features such as gender, age, work type, residence type, average glucose level, body mass index (BMI), and smoking status were consistently identified as key predictors. These results indicate that integrating swarm intelligence with ensemble learning can effectively improve stroke risk prediction and support clinical decision-making.
Abnormal behavior detection in crowded environments remains challenging due to complex motion patterns, occlusions, and domain variability. This paper presents transformer-based multi-modal attention network (TMA-Net), a unified framework that integrates red, green, and blue (RGB), optical flow (OF), and heat map (HM) modalities through a dual-stage attention fusion mechanism. The system employs you only look once version 11 (YOLOv11) for human localization and vision transformer (ViT)-B/16 for feature encoding, followed by intra-modal self-attention and cross-modal fusion to capture fine-grained spatial–temporal and motion energy dependencies. Extensive experiments on six public benchmarks as UMN, Crowd-11, UBNormal, ShanghaiTech, CUHK Avenue, UCSD Ped2, and EPUAbN dataset, demonstrate that TMA-Net achieves up to 97.5% area under the curve (AUC) and 96–100% accuracy, outperforming previous other state-of-the-art approaches. These results highlight the framework’s strong generalization and robustness across both single- and cross-dataset evaluations, underscoring its potential for reliable deployment in real intelligent surveillance systems.
Millions of people across the world are affected by hearing loss and early detection is very important for effective intervention. The traditional hearing screening methods are effective but they often rely on specialized equipment and clinical resources, making them less accessible to common people. Hearing loss is a state that affects the ability to communicate, socially interact and overall quality of life. The advancements in recent years have aimed to enhance the accessibility and efficiency of hearing tests, mainly in remote areas. The accurate classification of hearing loss is essential for effective detection and treatment in audiology. This study presents a deep learning (DL)-based approach based on a feedforward neural network (FNN). This paper focuses on common causes like cerumen impaction, otitis media, and otosclerosis. The study tries to explore ways to improve the diagnosis of hearing loss. The goal is to develop solutions that make hearing screenings more accessible and cost-effective for populations with limited access to healthcare resources. The results show the advantages of DL models in supporting automated accurate classification of hearing loss for intelligent diagnostic systems in audiological healthcare.
Microscopic image-based malaria detection still struggles to capture complex features due to variations in lighting and color. The support vector machine (SVM) method is often used in medical image detection, but its performance depends heavily on the selection of optimal kernel and hyperparameters (C and gamma). Conventional approaches, with single kernels and manual tuning, have limitations in capturing both spatial information and color distribution simultaneously. Therefore, this research proposes hybrid kernel support vector machine-cuckoo search algorithm (HKSVM-CSA) method that combines the radial basis function (RBF) kernel and histogram intersection for SVM, along with hyperparameter optimization using the CSA. The dataset used is malaria cell images, which contains parasitized and uninfected images of blood cells. The proposed method comprises five main steps: dataset preparation, feature extraction, HKSVM, hyperparameter optimization, and model evaluation. Experiments demonstrate that the proposed model achieves 94% accuracy, 93% sensitivity, 94% specificity, and area under the curve (AUC) of 0.98, which is significantly better than standard SVM, SVM-genetic algorithm (GA), and k-nearest neighbors (KNN). These results show that combining kernel and CSA significantly improves detection accuracy. This approach is promising for image-based automatic systems for infectious disease diagnosis.
Poses of human pattern recognition through machine learning is an essential facet of several applications, including health, surveillance, and sports analysis. For children yoga and kavayat (mock drill) increases the physical as well as mental health. Through the analysis of motion data this work discriminates between varied actions with high precision, contribution valued insights for monitoring and analysis in real-time. Here, this system that influences technique called as machine learning specifically logistic regression, to precisely discriminate and classify physical action patterns (child pose estimation) of children. Analogous to the propagation of false information, identifying physical actions is essential for upholding integrity and efficiency across various domains. The novel system demonstrates a towering accuracy of 98.00%, highlighting its effectiveness in recognizing and classifying physical action patterns.
