The number of people connected to internet has increased exponentially. This growth has spawned an unparalleled amount of information, both real and fake, that gets circulated through the digital networks with relative ease and permanence. The fast spread of fake news has created a desperate need of reliable and automated systems of detection. Ensemble learning has demonstrated some significant potential in alleviating this problem by boosting the strengths of classification by integrating diverse models. This study compares three ensemble learning methods - Bagging, Boosting, and Cascading to determine their relative effectiveness in classification tasks. The Bagging model, achieved an accuracy of 86.58% and a Matthews Correlation Coefficient (MCC) of 0.7392, showing high precision but low recall. AdaBoost with 50 estimators performed well with an accuracy of 83.74% and an MCC of 0.6752 in the Boosting category but Gradient Boosting outperformed it with an accuracy of 85.95% and a MCC of 0.7200. The cascading ensemble, combining multiple base classifiers, outperformed all other models with 90.06% accuracy, F1-score of 89.76% and Matthews correlation coefficient (MCC) of 0.8026. Findings reveal that cascading ensemble increases the model's resilience through the combination of complementary decision boundaries of different classifiers. Among the strategies that were evaluated, the cascading method with a multi model approach offers the best precision-recall balance, making it the most effective ensemble technique for the examined dataset.

Palm oil adulteration poses a major challenge to food quality, safety, and public health. Conventional analytical techniques, including chemical and spectroscopic methods, though reliable, are often resource-intensive and time-consuming. This work investigates the application of machine learning algorithms for predicting adulteration in palm oil using its physicochemical characteristics. K-Nearest Neighbors (KNN), Random Forest, Gradient Boosting, AdaBoost, Bagging, XGBoost, Support Vector Machine (SVM), Decision Tree, Naïve Bayes, and Logistic Regression were evaluated for the purpose of detecting palm oil adulteration. Their performances were evaluated using accuracy, precision, recall, and F1-score. Results indicate that ensemble-based methods, particularly Random Forest, Gradient Boosting, and AdaBoost, achieved superior outcomes with a maximum accuracy of 94.6% and consistently strong precision and recall values. These findings demonstrate that machine learning can provide a rapid, reliable, and cost-efficient alternative for detecting adulteration in palm oil.

An accurate assessment of soil fertility is crucial for enhancing crop production and for sustainable use of agricultural land, but conventional soil fertility evaluation methods can be time-consuming and labour-intensive. This research examines the use of Automated Machine Learning (AutoML), in particular the Tree-based Pipeline Optimization Tool (TPOT), to create a fast and accurate prediction model for soil fertility assessment. The study used a dataset soil samples from a soil testing laboratory and Soil Health Cards. The properties include important soil physicochemical, macronutrients, micronutrients and electrical conductivity. TPOT was used to automatically design machine learning pipelines with feature selection and hyperparameter tuning using genetic programming. The model was highly accurate (97.63% cross-validation accuracy) in predicting soil fertility. The research showcases the benefits of AutoML in terms of efficiency and performance. The results indicate that TPOT methods can assist in data-driven agriculture, and offer a scalable approach for soil fertility evaluation.

Crop classification is important in improving agricultural production, efficiency in resource management and sustainable agriculture. The paper introduces a machine learning-based scheme of intelligent crop classification using agro-climatic and soil-related variables that are highly interconnected with the definite crop growth needs. The dataset includes the most important attributes, including the nutrient content (N, P, K), temperature, humidity, soil pH, and rainfall, all of which affect the suitability of crops and the potential to produce. Numerical methods of operation have contributed immensely to the efficiency of crop recommender systems in the sense that they have been able to capture the complexity of interactions between environmental factors and crop reactions. Four machine learning algorithms, i.e., Random Forest Classifier, Gradient Boosting Classifier, K-Nearest Neighbors, and Support Vector machine, were adopted and tested using typical performance indicators of accuracy, precision, recall and F1-score. In the experiment, the highest accuracy of 0.948 was obtained with the help of Random Forest and Gradient Boosting classifiers, which proves that they are more effective in approximating non-linear patterns and minimizing classification error. K-Nearest Neighbors and Support Vector machine yielded good results also having a 0.943 and 0.936 accuracy respectively and this confirms their applicability in the differentiation of crops using multi-dimensional feature space.

Parkinson's disease (PD)—being a gradual neurodegenerative disorder affecting motor and cognitive skills—requires early and proper diagnose for effective management. This work applies advanced ensemble machine learning methods, stacking, and cascading algorithms for a better diagnosis of Parkinson's disease. The study uses a diverse dataset consisting of patients' demographic data, clinical features, and diagnostic features, trying to prove that ensemble methods outperform individual machine learning algorithms. The results show that ensemble learning methods much better predict and resist uncertainty, proving better than the base classifiers with 95.01% accuracy, 96.93% recall rate, and 0.8938 in Matthews Correlation Coefficient (MCC). That is to say, the other methods do offer improvements la stacking, which forms a meta-classifier out of multiple models, and cascading, which processes predictions from one model into the next, gives great improvements over stand-alone methods with very strong diagnostic precision. Such findings suggest that ensemble learning effectively minimizes false classification rates and thereby improves the early detection of Parkinson’s disease. The study emphasizes machine learning, especially ensemble methods, as the operating system within decision support for clinical systems for neurodegenerative disorders. With the diversity in patient data and model integration optimization, they certainly pave the road to enhanced diagnostic accuracy. Subsequent investigations may consider the use of larger multicenter data sets, more biomarkers, and deployments in real-time healthcare settings to further assess and improve predictive models. This document thus adds to an ever-increasing body of research in support of AI applications within neurology and personalized medicine.

Most of the developing countries rely on agriculture for their annual growth, a greater section of GDP contribution is affected by its agriculture. Alongside this, major industries rely upon the agricultural to derive their profit. The farmers of many countries are still dependent on classical methods as they are not aware of various technologies and lack significant knowledge in the sector. Consequently, the loss in the agricultural domain accounts for poor quality seed and delayed sowing, environmental hazards, attacks by insects, irrelevant harvesting etc. However, these factors cannot be controlled but can be monitored and an effective strategy can be adopted with the help of the Internet of Things (IoT). IoT-assisted agriculture can help to monitor the conditions and predict the yield which can be useful in taking decision about crops. However, there are no concurrent real-time forecasting strategies that can avail the inputs from different sensors and give parameter-specific or complete forecasting of the yields. To overcome this issue, a Rule-Based Context-Aware (RBCA) framework is proposed that considers a specialized IoT setup to help yield-forecasting of different crops by using available parameters as input. The proposed model uses a probabilistic mapping theory with residual analysis. The model can help to analyze if the given conditions of agriculture-setup are good enough for the growth of a particular crop by using the standard growing environment as matching criteria. Both numerical and realtest data for various crops between the years 1981 and 2012 are used for evaluating the performance of the model.

Locating accurate lung cancer classifications remains essential for proper medical treatment because lung cancer maintains its position as the primary source of cancer-related deaths. This study presents a comparative analysis of deep learning approaches, specifically ResNet50 and EfficientNetB0 for the classification of lung cancer using histopathological images. By leveraging the balanced LC25000 dataset containing three categories (adenocarcinoma, squamous cell carcinoma, and benign), the research implements comprehensive preprocessing, data augmentation, and fine-tuning of transfer learning models. Performance evaluation based on metrics such as accuracy, precision, recall, F1-score, and ROC-AUC demonstrates that EfficientNetB0 surpasses ResNet50, achieving near-perfect classification (overall accuracy 95.93%) and perfect AUCs across all classes, while ResNet50 shows pronounced misclassification, especially for adenocarcinoma (overall accuracy 34.63%). These findings underscore the superior suitability of EfficientNetB0 for practical deployment in lung cancer histopathological image analysis and highlight the need for additional optimization in traditional architectures to improve diagnostic reliability.

The adoption of artificial intelligence (AI) in healthcare has led to breakthroughs in disease diagnosis, treatment planning, and patient monitoring. However, the black-box nature of many AI models poses a significant barrier to clinical adoption. Explainable AI (XAI) offers a solution by making model decisions transparent and interpretable to healthcare professionals. This systematic review examines the state-of-the-art XAI techniques applied in healthcare, categorizing them by model-specificity (intrinsic vs. post-hoc), data type (images, text, tabular), and clinical application. We analyze 125 peer-reviewed studies published between 2018 and 2024 across key medical domains. Our findings highlight trends, challenges, and future directions, emphasizing the need for user-centric and domain-specific interpretability. This review serves as a comprehensive reference for researchers and clinicians seeking to implement interpretable AI solutions in healthcare.

This study evaluates the performance of VGG19 and VGG16 deep learning models for diagnosing Parkinson's disease (PD) using biomedical data. The objective was to assess their diagnostic accuracy and suitability for clinical applications. Both models were trained and tested on a dataset, with VGG19 achieving a test accuracy of 88.51% and an F1 score of 89.44%, while VGG16 recorded 86.49% accuracy and an F1 score of 87.95%. VGG19 slightly outperformed VGG16 in classifying PD-positive cases, with both models showing minimal overfitting and strong generalization on unseen data. Confusion matrix analysis confirmed balanced precision and recall, though potential class imbalances in the dataset require further investigation. These findings highlight the efficacy of VGG19 and VGG16 for early PD detection, emphasizing their potential to support timely interventions. Future improvements could include data augmentation, transfer learning, or threshold tuning to enhance diagnostic specificity. While promising, validation on diverse datasets is essential before clinical adoption.

This study considers the use of machine learning techniques to predict kidney disease and its progression, given a rich dataset that includes clinical and demographic features. compared with random forest, gradient boosting, and logistic regression as well as advanced ensemble techniques like stacking and cascading. The data was preprocessed well such as one-hot encoding, scaling and no missing values, in order to allow robust training of the model. Performance of Random Forest was 0.80, 0.64, 0.80 and 0.71 for accuracy, precision, recall and F1-score, respectively, and similar results in the remaining models. Potassium and blood urea levels emerged as important risk predictors in the Random Forest models, as revealed by extensive visualizations (e.g., ROC plots, feature importance plots, correlation heatmaps, PCA projection in 3D). The novelty of this work is to apply the state-of-the-art ensemble learning methods, in particular stacking and cascading, to kidney disease prediction, an application that has not received extensive attention in this domain. Moreover, by integrating prediction and interpretability via visualization and feature exploration, this work not only enhances diagnostic performance but also discovers biologically relevant biomarkers. Future work will focus on hyperparameter optimization, feature engineering, and empirical validation to enhance model generalizability for real-world clinical applications

Accurate fracture classification is essential for efficient diagnosis and treatment planning. This paper presents a CNN-based automated fracture classification model and verifies its high accuracy and reliability. The model was verified with a confusion matrix heatmap, which verified its exceptional performance with negligible misclassifications. Fracture Dislocation, Comminuted Fracture, and Pathological Fracture were correctly identified with 100% accuracy, and misclassification errors were insignificant. The model achieved an overall accuracy of 99%, and precision, recall, and F1-scores were always between the range of 0.97 and 1.00. High macro-average and weighted average scores also validate its generalizability. The CNN architecture, consisting of convolutional, pooling, and fully connected layers, efficiently extracts and processes useful features in a computationally efficient way. Training analysis indicated a high convergence trend, with accuracy progressively improved and loss substantially decreased with each epoch. The steep decline of loss in initial stages of training verifies an optimized learning process, which testifies to the reliability of the model. This paper suggests that the proposed CNN model offers a strong and efficient solution for automated fracture detection. Future work will involve dataset expansion, including heterogeneous imaging modalities, and model optimization for clinical deployment.

Parkinson's disease (PD) is a chronic, non-fatal, and well-known progressive neurological disorder, the symptoms of which often overlap with other diseases. Effective treatment of diseases also requires accurate and early diagnosis, a way that patients can lead healthy and productive lives. The main PD signs are resting tremors, muscular rigidity, akinesia, postural instability, and non-motor signs. Clinician-filled dynamics have traditionally been an essential approach to monitoring and evaluating Parkinson's Disease (PD) using checklists. Accurate and timely diagnosis of Parkinson's disease (PD), a chronic and progressive neurological ailment, can be difficult due to its symptoms overlapping with those of other disorders. Effective therapy and improvement in the quality of life for patients depend on early and accurate detection. To improve classification performance, this study investigates transfer learning, which uses pre-trained models to extract features from massive datasets. Transfer learning improves generalization and permits domain adaptation, especially for small or resource-constrained datasets, while lowering training time, resource needs, and overfitting concerns. This work aims to design and assess a general transfer learning paradigm for the reliable prognosis of Parkinson’s disease based on DaTscan images that consider feature extraction and the performance of a variety of ML algorithms. This work aims to explore the use of transfer learning with pre-trained deep learning models to extract features from DaTscan images in order to improve classification accuracy. The sample of this study is made up of 594 DaTscan images from 68 participants, 43 with PD and 26 healthy. Out of the four algorithms employed; the Random Forest, Neural Network, Logistic Regression, and Gradient Boosting models, transfer learning-based features were applied. Four indices of accuracy, namely Area Under the Curve (AUC), Classification Accuracy (CA), F1 Score, Precision, Recall and Matthews Correlation Coefficient (MCC) were used to evaluate four machine learning models on a PD classification task such as Random Forest, Neural Network, Logistic Regression, and Gradient Boosting. Neural networks outperformed the other models, showing robustness and reliability with an AUC of 0.996, CA of 0.973, and MCC of 0.946. Gradient Boosting performed competitively, coming in second with an AUC of 0.995 and MCC of 0.925. Random Forest performed the worst, with an AUC of 0.986 and an MCC of 0.905, whereas Logistic Regression had an AUC of 0.991 and an MCC of 0.926. These results demonstrate how well neural networks perform high-precision tasks and point to gradient boosting as a more computationally effective option.

Retinal diseases need to be recognized early and precisely in order to avoid unnecessary blindness. The paper is a comparative analysis of two popular deep learning architectures, EfficientNet and ResNet-50, which are used as feature extractors with four additional classifiers: Neural Network, Logistic Regression, Gradient Boosting, and Random Forest. Experimentation was done on a dataset of 4,217 fundus images (4 eye conditions, Normal, Cataract, Glaucoma, and Diabetic Retinopathy). The deep features were extracted and they were used to train the classifiers and their performance measured by using accuracy, F1-score, AUC and MCC. It is observed that EfficientNet with a Neural Network gives the best results with an AUC of 0.980 and a precision of 0.888. Logistic Regression and Gradient Boosting also score well with the Random Forest always scoring the lowest points. The same tendency is observed in ResNet-50 features, as once again, the Neural Network works better than the rest of the classifiers. The results indicate the significance of a proper choice of a classifier alongside a powerful feature extractor, and reveal how the hybrid deep feature-classical classifier methods can be used to improve the diagnosis of retinal diseases to a considerable degree.

Agricultural productivity remains a crucial element of food security and economic stability in many countries. With climate variability, land degradation, and increasing demand for precision agriculture, accurate crop yield prediction has become essential for informed decision-making. This examines ambitions to broaden and examine a system learning (ML) model for predicting crop yields using environmental, soil, and ancient yield information. The research utilized a dataset combining meteorological variables, soil properties, and crop control practices throughout 3 Indian states over ten years. Several ML algorithms, which includes Linear Regression, Decision Tree Regression, Random Forest, and XGBoost, have been evaluated. The Random Forest model outperformed others with an R² rating of 0.89 and RMSE of 2.Five quintals/ha. The consequences demonstrate the capability of ensemble mastering models in dealing with non-linear agricultural statistics. The take a look at concludes with the consequences of ML-based yield forecasting structures for precision agriculture and coverage-making.

Soil fertility is a critical factor in agricultural productivity and sustainability. Accurate soil fertility assessment can lead to better crop management and yield optimization. This paper presents a data-driven approach to soil fertility assessment using an Extreme Learning Machine (ELM), an advanced machine learning algorithm known for its fast-learning speed and high generalization capability. The study involves comprehensive exploratory data analysis (EDA) and feature engineering to identify the most relevant soil properties influencing fertility. A dataset comprising various soil attributes was preprocessed, standardized, and subjected to feature selection techniques such as Recursive Feature Elimination (RFE) and Principal Component Analysis (PCA). The ELM model was trained on this refined dataset and evaluated for its performance in predicting soil fertility ranks. This research highlights the potential of using ELM for soil fertility assessment, offering a robust and efficient tool for agricultural stakeholders. The findings can contribute to more informed decision-making in soil management and crop planning, ultimately enhancing agricultural productivity.

Hair and scalp diseases represent a significant medical challenge, affecting millions worldwide. Early and accurate detection of these conditions is crucial for effective treatment and management. The proposed system integrates advanced DL algorithms with image processing methodologies to analyze digital images of the scalp and hair. By leveraging a diverse dataset of annotated images, the system employs supervised learning techniques to train models capable of recognizing patterns indicative of different hair and scalp diseases. Feature extraction, segmentation, and classification algorithms are applied to process and interpret image data, facilitating accurate diagnosis and classification of conditions such as alopecia, dandruff, psoriasis, and fungal infections. In this research, a deep learning technique was employed to accurately forecast ten major categories of hair loss and scalp disorders, including alopecia, psoriasis, and folliculitis. However, the study faced challenges due to limited prior research, a lack of suitable datasets, and the diverse nature of images available online. To address these challenges, 9600 images were used for training and 1200 for testing, with preprocessing steps such as denoising, image equalization, enhancement, and data balancing to minimize the error rate. The preprocessed data was then fed into a convolutional neural network (CNN) model, resulting in an overall training accuracy of 99.2% and a validation accuracy of 98.1%.

Gestures have a vital role in the selection procedure of an interview. Kinesics, which includes expressions of the face, hands, head, eyes, body posture, and appearance of the candidate, is a vital parameter for a job interview. To keep this view in mind, the researchers in this paper have explored and identified the positive and negative gestures that are essential for a candidate to know before appearing in an interview. The researchers in this study employed machine learning technology that can be utilized to understand human conduct during mock interview sessions in universities and colleges for students preparing for placement in alleged companies. It relates to the recognition of expressions of motion involving the movement of hands, face, eyes, and head. The researchers in the present paper have employed different machine-learning techniques like “Random Forest”, “Neural Network”, “Logistic Regression”, and “AdaBoost” to measure human behavior in the framework of placement discussions.

Diabetes is a chronic disease that affects millions of people worldwide. Early detection and timely intervention are crucial for preventing severe complications. Traditional diagnostic methods can be time-consuming and invasive. This research explores the application of machine learning techniques to develop a non-invasive and efficient diabetes screening system. By analyzing various patient data, including demographic, clinical, and lifestyle factors, the model aims to accurately predict the likelihood of developing diabetes. The study compares the performance of different classification algorithms, such as Gradient Boosting, Random Forest, and Logistic Regression, in terms of accuracy, precision, recall, and AUC-ROC. The results demonstrate the effectiveness of Gradient Boosting in accurately identifying individuals at risk of diabetes. This research has the potential to significantly improve early diabetes detection and contribute to better disease management and prevention.

Crop prediction has historically been influenced by farmers’ needs, but the impact of climate change on crop yields has made traditional methods unreliable. Factors such as soil quality and environmental conditions are no longer sufficient for determining which crops to grow. Manual predictions have proven ineffective, prompting the need for accurate crop prediction methods. Machine learning has emerged as a crucial tool for this purpose, considering various factors like soil composition, geographical location, and climate conditions. Selecting the appropriate crop based on these attributes is essential for accurate predictions. This paper examines multi-class crop prediction models employing machine learning techniques. Specifically, Xgboost and Random Forest classifiers were utilized in our analysis. While both models demonstrate good accuracy, The findings of this study indicate that both Random Forest and Xgboost models achieve relatively high accuracy rates in predicting crops from the dataset. Specifically, the Xgboost model outperforms the Random Forest model slightly, with an accuracy of 83.25% compared to 82.5%.

Currently, the move from traditional healthcare to smart healthcare systems is greatly aided by current technology. Healthcare proposes a new healthcare model that is patient-centered using advancements in wearable sensors, connectivity, and the Internet of Things (IoT). The administration of enormous amounts of data, including reports and pictures of every individual, increases human labour requirements and security hazards. This study shows how a blockchain-based Internet of Things might improve patient care while lowering costs by using medical resources more wisely. Initially, Resource Provider’s IoT data will be sensed and encrypts using Diffie Hellman Galois–Elliptic-curve cryptography (DHG-ECC). Next, from the extracted attributes, the optimal features will be selected by using Pearson Correlation Coefficient based Sand Cat Optimization Algorithm (PCC-SCOA). After that, the selected optimal features will be combined and converted into hashcode using the Digit Folding–Streebog Hashing algorithm. This hashcode will be constructed in the form of Smart Contract. Next, the Resource Requester (Doctor or Nurse) sends the Role Request with the Combined Linear Congruential Generator–Digital Signature Algorithm (CLCG-DSA). The next Resource Requester will be matching the hashed access policy with Blockchain. The proposed models are used to compare the performance of proposed design using feature selection time, Encryption time, Decryption time, security level, signature creation time and signature verification time. Our proposed method DHGECC approach achieves 96.123