Advanced Computing

Plant diseases are a major threat to agricultural production globally, resulting in decreased crop yields and financial difficulties. For these illnesses to be effectively managed, early and precise disease identification is essential. Through the use of both deep learning and conventional machine learning techniques, this work proposes a thorough method for classifying plant leaf diseases. The research makes use of a library of tagged plant leaf photos that includes both healthy and diseased leaves. The leaf photos are first processed using AlexNet, a deep convolutional neural network (CNN), to extract complex characteristics. The dataset is utilized to train the CNN model, and its high-level feature representations are applied to categorize diseases. For comparison analysis, classic machine learning techniques like Naive Bayes (NB) and K- Nearest Neighbors (KNN) are also used. To test these algorithms’ ability to identify between various plant diseases, they are applied to the derived characteristics. In the context of classifying plant diseases, the comparative study attempts to assess the benefits and drawbacks of both deep learning and traditional machine learning methodologies. The findings of this study offer insightful information about the effectiveness of several plant disease diagnosis methods. A multifaceted strategy to reliably diagnose plant diseases is provided by the integration of deep learning and machine learning techniques, assisting farmers and agricultural specialists in making timely disease management decisions. This study contributes to continuing attempts to lessen how plant diseases affect the sustainability of agriculture and the safety of the world’s food supply.

The Internet of Things (IoT) and the Internet of Vehicles (IoV) represent cutting-edge technologies with the potential to significantly impact various sectors, particularly disaster management and public safety. This research introduces an innovative system for real-time earthquake and fire detection, strategically deploying sensors in vehicles and buildings. These sensors collect data transmitted to a central unit, enabling rapid detection through advanced analytics. The system leverages IoV technology for quicker emergency responses and efficient evacuation routes. Compared to traditional systems, it offers faster disaster response, reducing risks to lives and property. The use of IoV technology enhances decision-making based on real-time traffic data, increasing adaptability. This study explores the implementation and potential impact of this IoT and IoV system on public safety and disaster management.

As the Global population ages, protecting the welfare of the elderly becomes a more pressing issue. The prompt diagnosis of falls, which are a major cause of injuries and fatalities among older individuals, is a crucial component of geriatric care. Early fall detection (EFD) systems are essential for giving prompt help and raising the standard of living for elderly people. Traditional fall detection algorithms often suffer from false positives, where non-fall events are incorrectly identified as falls, or false negatives, where actual falls are missed. Hence, researchers and developers are increasingly turning to more sophisticated machine-learning techniques to improve the precision and reliability of systems used for fall detection. Advanced machine learning approaches are being used to improve these systems’ accuracy and effectiveness, and one approach that is gaining popularity is the knowledge distillation ensemble. In this paper, we propose early fall detection in elderly people using the knowledge distillation ensemble (KDE) method to ameliorate the reliability and accuracy of the advanced machine learning approaches. We conducted experiments using our proposed method to detect falls using physiological parameters and we evaluated our work using metrics like accuracy, F1-measure, recall, and precision. Our proposed KDE algorithm has achieved 100% accuracy and the perfect score for precision, recall, and F1-measure.

In the agricultural domain, sugarcane crops, like many others, are susceptible to diseases, posing a significant threat to both quality and quantity of production. Identifying and mitigating these diseases in their early stages are critical to averting financial losses for farmers. In response, researchers have turned to Artificial Intelligence (AI) techniques such as Machine Learning (ML) and Deep Learning (DL) to analyze diverse agricultural data, including yield prediction, climate patterns, and soil quality, with disease prevention being a prime focus. This paper presents a thorough exploration of the effectiveness of a Deep Learning-based Convolutional Neural Network (CNN) algorithm tailored for the detection of prevalent sugarcane diseases in India. Motivated by the rapid evolution of disease classes and farmers’ limited diagnostic skills, this study employs advanced deep learning and computer vision techniques. Through image categorization into healthy and diseased groups, the trained model achieves an impressive 98.69% accuracy rate in sugarcane disease detection. Furthermore, to empower farmers, a web-based application is developed for ongoing disease monitoring. The paper suggests future research avenues, including user feedback integration and exploring the intersection of disease detection with agricultural productivity enhancement and price forecasting, thus enriching farmers’ decision-making processes.

Economy and financial status of a country is largely dependent on agriculture. Virtual Assistants (VA) are successfully deployed in different applications, however its use in the farming domain is very limited. There are number of challenges in this like lack of single database which will satisfy the varied needs of farmers like crop prediction, disease management, prediction about proper harvest time, locating nearby godowns, weather forecast etc. Along with this one of the major problems is limited technology literacy of the users poses difficulty in designing and deployment of VA in farming domain. Many times, it is difficult to describe the field conditions verbally. In this case it will be better to accept the photo-based input from the user. The user can pass on additional queries using audio input. By this method the effectivity of VA can be increased. In this work the plant disease prediction is designed with the help of plant leaf images. The system is trained on images taken from PlantVillage Dataset. The system is trained on 61 images of 5 different class. The database contains the images from various crops like maize, strawberry, tomato. 12 different features of diseased segment are used to have the trained model. It is found that the accuracy of SVM classifier is 88.5% and using NN the accuracy is improved to 90.16%. To improve the accuracy further, the image datasize should be increased. Deep learning algorithms like EfficientNet, InceptionNet, ResNet,etc. have better accuracy. In future we plan to integrate this with a system that will generate the audio responses for audio queries raised by farmers.

In the seafood business, fish freshness plays a crucial role since it directly affects quality, customer happiness, and safety. This study uses a well-selected dataset of fresh and non-fresh Tilapia fish species to assess the performance of several CNN models, including VGG-19, MobileNetV2, DenseNet201, and ResNet50, for classifying fish freshness. DenseNet201 performed exceptionally well with an accuracy of 1.0, MobileNetV2 had a high accuracy of 0.99104, VGG19 performed admirably with an accuracy of 0.964809, and Resnet50 offered competitive accuracy of 0.82098. To achieve these results, we designed and implemented a rigorous procedure for training and testing these CNN models using the dataset of both fresh and non-fresh fish species. We used meticulous data preprocessing and model training, bearing in mind the importance of high-quality datasets. Our study’s primary findings emphasize how crucial it is to select an appropriate CNN architecture and take dataset quality into account when determining the freshness of seafood. Concerning fish freshness evaluation, DenseNet201 and MobileNetV2 in particular demonstrated remarkable accuracy, underscoring the importance of model selection and data quality.

Deep learning is being used a lot to develop a quick, automatic and reliable means for image identification and classification in many domains. Using Deep learning techniques in the agriculture would be an enhanced practice in the history of agriculture. Chillies are one of the most popular crops in India as they are used in everyday life for cooking variety of dishes. Chilli plants are sensitive to multiple infections. Detection and prevention of these diseases to other parts of the plant is a very important and impracticable task in the case of large fields. This paper proposes a disease detection and classification model for chilli leaves using Convolution Neural Networks. And also, other pre-trained architectures like ResNet, Inception, VGG (Visual Geometry Group) and Efficient Net were used for building an optimized model which detects the diseases more accurately. Images of Chilli leaves, having various diseases named Leaf curl, Leaf spot, Yellowish, as well as Healthy leaves, from the self-made dataset were used. The results of the Efficient-Net model prevailed over other models with accuracies of CNN (Convolutional Neural Networks), ResNet, VGG and Efficient Net 70%,87%,87% and 91% respectively.

Natural catastrophes including flooding, tornadoes, earthquakes, and wildfires have been occurring more frequently over the past few decades as a result of global warming and climate change. Therefore, it is more crucial than ever to give emergency response workers accurate and timely information to enable them to respond to crises effectively. Among the many pieces of information required for disaster response and management, it is crucial that rescue workers are promptly notified of the location and extent of a building's destruction in order to maximise the effectiveness of their efforts. Nevertheless, despite significant efforts, problems with picture classification for disaster response still exist. In this study, a potential deep learning-based method is put forth for identifying damaged buildings in high-resolution satellite photos. It solves the issue of limited training data common in many remote sensing applications by using generic data augmentation. It is suggested that a pretrained model be used in conjunction with transfer learning as a fine-tuning method for the relevant task. The trials with images of Port-au-Prince, Haiti showed that the suggested strategy works well with sparse training data. With enriched training data, the Convolutional Neural Network (CNN) model can detect damaged buildings with an accuracy of 83%, compared to only 53% with the original training data. The focus of future study will be on investigating automated ways to obtain larger training datasets and model generalisation by researching more reliable data augmentation strategies.

Our goal is to determine the parameters that affect the total electron content in the ionosphere (TEC) by comparing data with numerous models. Free charged particles are present in the plasma of ionised gas that makes up the terrestrial ionosphere. It is created when solar radiation ionises. IRI is present in the Earth's atmosphere and is a component of gaseous elements. The magnetosphere's dense ions and charged particles have an effect on the speed of radio-frequency signals. Therefore, one of the most significant causes of inaccuracy in GNSS (Global Navigation Satellite System) positioning and navigation services is magnetospheric delay. Furthermore, the ionosphere's quantitative influence clarifies the total electron content (TEC), which is the total number of electrons gathered per square metre during the journey from a spacecraft to a GNSS receiver. We are attempting to determine the relative performance of various machine learning techniques, including Gradient Boosting Model, LSTM, and Linear Regression, on the TEC prediction problem. The experimental investigation demonstrates that the gradient boosting regressor produced the minimum loss followed by a legitimate coefficient of determination when comparing all models.

In current scenario of healthcare, Diabetes Mellitus stands as an incurable condition, underscoring the imperative of early detection. Factors contributing to the onset of diabetes encompass aging, weight gain, sedentary lifestyle, genetic predisposition, poor nutrition, irregular routines, elevated cholesterol levels, and other associated conditions. The intersection of healthcare and machine learning unveils intriguing possibilities, capturing the attention of medical professionals. This study aspires to empower healthcare practitioners in predicting diabetes at an early stage through the application of machine learning techniques. By scrutinizing and comparing classification algorithms, including Random Forest, Supervised Machine Learning, and Decision Tree, we sought to discern their efficacy in forecasting diabetes mellitus. A systematic evaluation identified a model achieving an impressive accuracy rate of 98.56%, offering a substantial contribution to the utilization of machine learning for diabetes prediction. This research augments our understanding of the practical implications of machine learning in the healthcare domain, particularly in the context of early disease detection.

Diabetes, a chronic metabolic disease with a rising global prevalence, significantly impacts individuals’ health. Diabetes increases a person’s risk of developing various diseases, including heart disease, stroke, vision problems, nerve damage, etc. Early detection and proactive care of diabetes can lessen its impact and improve patient outcomes. Utilizing the powers of machine learning algorithms in the medical field has shown significant promise in accurately identifying diseases and implementing customized treatments, reducing the workload of healthcare professionals. This paper proposes a methodology based on Gradient Boosting technique to accurately predict diabetes. This study also provides a thorough analysis of diabetes prediction using a variety of classifiers, including Linear Discriminant Analysis (LDA), Extra Tree Classifiers (ETC), Quadratic Discriminant Analysis (QDA), Stochastic Gradient Descent (SGD), Bayesian Gradient Descent Classifiers (BGC), and Gradient Boosting (GB) classifiers. The pre-processing methods of Standard Scalar Normalization and Synthetic Minority Over-sampling Technique (SMOTE) are used to improve the predictive models’ quality. SMOTE is used for class balancing. Accuracies achieved by LDA, ETC, QDA, SGD, BGC, and GB are 77.34%, 74.20%, 73.50%, 75.01%, 74.08%, and 80.19%, respectively. The authors optimized the Gradient Boosting (GB) classifier through a rigorous grid search optimization process to maximize performance, yielding an accuracy of 82.70%.

This article introduces a method for classifying diabetes based on machine learning (ML) methods. In recent years, significant focus have been put onto increasing disease classification performance through the use of ML approaches. This paper outlines the use of five interpretable ML algorithms: Bagging classifier, Random Forest, AdaBoost, Multilayer Perceptron, and Restricted Boltzmann Machine. All the ML classifiers were trained and tested in a benchmark Biostat Diabetes Dataset using Python programming. Each technique’s performance is evaluated to discover which has the finest accuracy, precision, recall, F1-score, specificity, and sensitivity. Experimental findings and assessment reveal that the Random Forest technique outperforms all other ML techniques by achieving 98% precision, 98% recall, 98% F1-score, 75% sensitivity, 96% specificity, and accuracy of 97.5%.

Vital signs indicators like temperature, heart rate, and oxygen saturation should be examined periodically, as these are the root cause of any medical diagnosis. Any deviations from the normal range indicate that the person needs an immediate medical check-up and is on the edge of facing some medical issues later. Thus, monitoring these vitals periodically can help patients from the risk of mortality. The goal of this research is to forecast a person’s abnormality using machine learning and IoT (Internet of Things) algorithms for decisiveness. The prototype was built using three sensors, MAX30100 sensor (for SpO2), REES52 heartbeat sensor, and LM35 temperature sensor along with Arduino UNO, and ESP 8266. The bio signal data from these sensors were collected using Arduino UNO, stored in a local PC, and uploaded to the cloud using API protocol in Thingspeak (IoT platform). These data were also retrievable for further diagnosis. Support vector machine (SVM), a machine learning method, is used to predict if a patient is abnormal or not. SVM learns the threshold ranges for each parameter as well as the associated goal value from the datasets.

Various advanced computing techniques and capabilities have the deep impact in the field of medical sciences, especially in identifying human heart diseases. So that identifying heart related diseases accurately and in time may save the patients life’s and increases the chances of survival. However, manual approaches for identifying heart disease suffer from biases and variations between examiners. We can use various machine learning algorithms to overcome these issues in manual approaches These ML algorithms provide more accurate and efficient tools for identifying and analysing the patients with heart disease.To explore the potential of machine learning algorithms, the recommended study employed various techniques to identify and predict human heart disease using a comprehensive heart disease dataset. Sensitivity, specificity, F-measure and accuracy in classification can be used to examine the performance. We have used eight machine learning classifiers such as Ada boost, Extreme Gradient Boosting including Decision Tree, Logistic Regression, Linear Discriminate Analysis, Random Forest, Naïve Bayes, Support Vector Machine. The results demonstrated notable improvements in the prediction classifiers’ accuracy. This underscores the efficiency of machine learning algorithms in finding and predicting human heart disease. This research achieved improved accuracy heart disease prediction using the machine learning technique. Multiple classifiers were employed to classify heart disease prediction, with SVM achieving an accuracy of 95.88%.

Alzheimer’s Disease constitutes one of the biggest portions of the diseases related to ageing. Mild Cognitive Impairment may be considered the formative stage of this disease. The automated diagnosis of Mild Cognitive Impairment using Machine Learning will help the clinicians in delaying its progression and will be easy, cheap, and efficient for the patient. This work uses neuropsychological data obtained from Alzheimer’s Disease Neuroimaging Initiative (ADNI), containing the results from 12 tests including Mini-Mental State Examination and ADAS-Cog. An extensive empirical analysis is carried out and the most important features are extracted using the proposed pipelines. The Feature Selection is done using both filter and wrapper methods and in total 13 features were selected. It was found that most of the selected features related to tasks associated with memory. The proposed method gives a performance of 0.9817 in terms of F1 Score. Thus, performing better vis-à-vis the state of the art. The proposed pipeline helps to reduce the number of neuropsychological tests to diagnose the disease. This work is one of the components of the projects that use multi-modality data including structural-Magnetic Resonance Imaging, functional-Magnetic Resonance Imaging, Positron Emission Tomography and Neuropsychological data to develop a system for efficient and effective diagnosis of MCI. The project management is done using Agile Methodology. The results are encouraging and pave the way for the development of such a system.

The term “heart or cardiovascular disease” is frequently used to refer to a variety of heart-related issues. It is one of the illnesses with the highest mortality rate. Its fatality rate results in nearly 17 million deaths worldwide. Given the daily increase in heart disease cases, predicting any disease early is critical and concerning. Early detection of heart disease is a difficult task that must be accomplished carefully and effectively by gathering precise information about the patient who is at risk for heart disease. To predict whether a person has cardiac disease, there are numerous AI-based (Machine/Deep learning) Models available considering medical characteristics (features) of heart disease. By comparing and contrasting several machine learning classifiers, this paper proposed a heart disease prediction system. This system will aid in identifying and predicting whether the patient is likely to have or be diagnosed with heart disease based on his or her current medical state.

Artificial intelligence (AI) has the potential to transform health care as it has revolutionized many pattern recognition applications. During the last few years, medical image analysis has been gaining attention. Research on medical images using machine learning (ML) has made significant progress. The purpose of this study was to compare the accuracy of classification in clinical images among ML algorithms. Based on features extracted technique local binary patterns (LBP), histograms of gradients (HOG) and pixels feature extractor, seven classification models are compared. Several methods are used to classify important features are obtained by different feature extractors, including support vector machines (SVM), decision trees (DT), logistic regression(LR), random forests (RF), extreme gradient boosting (XGB), K-Neighbors classifiers (KN) and multinomial Naive Bayes (NB). To test the accuracy of our classification and feature extraction models specifically for histopathology images, we used COVID-19 chest radiographs, which is available publicly dataset containing 21,212 CT images divided into four classes. In comparison to other feature extractors, SVM has the better result using HOG as features. LPB feature extraction has been shown to be superior when used with SVM algorithm to classify COVID-19 chest radiograph data, as demonstrated by experiments on COVID-19 chest radiograph data.

Parkinson syndromes are a group of disorders affecting the elderly population with unsteadiness, slowness of activities, frequent falls, and speech disturbances, which slowly progress. Diagnosis of this group of syndromes is usually purely clinical and could be delayed due to its varied presentations. Parkinson’s syndromes comprise of Idiopathic Parkinson’s disease, Multiple system atrophy (MSA), progressive supranuclear palsy (PSP), and cartico basal ganglionic degeneration. In this work, we provide a comparative analysis of several deep learning models such as ViT, MobileNetV2, DenseNets, ResNets, GoogLeNet, VGGs for the differentiation of Parkinson’s syndromes using prolonged vowel phonations. To address this multi-class classification problem, we employed transfer learning on DL models, by training on a dataset comprising 337 sustained vowels from patients with parkinson’s disease, MSA, PSP, and no parkinson syndromes. Each recording is transformed into a mel-spectrogram for input into the models. Among the models, ResNet152 outperformed the other models, achieving an impressive accuracy of 98.30% in classifying parkinson disorders, offering a promising non-invasive, and cost-effective diagnostic tool for early intervention and treatment planning.

Brain tumors are fatal worldwide and are difficult to treat. The process requires time and is prone to error for medical professionals to examine the scans and identify tumor locations. To overcome the limitations, an efficient tumor detection and classification method is necessary for obtaining robust features as well as perform proper disease classification. This paper proposes a multiclass brain tumor classification based on hybrid Squeeze-and-Excitation Networks (SENET) with K-Nearest Neighbour (KNN) Algorithm. To improve poor contrast and raise the quality of the input images, the proposed design gathers and pre-processes the MRI images of brain tumors using the Recursively Separated Exposure Based Sub-Image Histogram Equalization (RS-ESIHE) technique. Following image enhancement, these images are given into FANET segmentation method to segment based on feedback mechanism during training and then using a hybrid SNET with KNN classification technique extracted significant features and classified brain tumor types. Accuracy, F1_score, precision, sensitivity and kappa are some of the metrics used to measure performance, and the results are 97.5%, 94.74%, 95.16%, 94.74% and 96.53%. As a result, the experimental findings for the proposed technique are superior to those of the other existing methods.

Psychological health problems concern an approximated 92 million people universally. That's essentially 1 in 10 people in general. Therefore, it is advisable to create a chatbot to lessen the stigma as-sociated with mental health, giving people the ability to voice their problems, and filling the gap left by the lack of support systems for those who need assistance. The preceding few years have been challeng-ing for everyone all around the world. The global escalation of Covid-19 has resulted in a substantial rise in the number of people suffering from emotional health problems. As a result, people are becoming more conscious of psychological wellness because a single consultation with a psychiatrist is expensive to execute, ideas are introduced for patients to be informed of their illness before scheduling an appoint-ment. The rise in mental diseases caused by loneliness and stress inspired us to create Mindful. The bot is a tool that allows people to converse in real time using developed rule sets or the assistance of simulat-ed intelligence and machine learning.

Internet of Things (IoT) and machine learning technology integration has had a significant positive impact on contemporary healthcare systems. The main objectives of this project are to develop and evaluate an integrated healthcare system based on the Internet of Things for the diagnosis and treatment of kidney-related illnesses. The system, which also uses a variety of sensors to continuously track essential health data, enables real-time communication between patients and medical professionals. Five machine learning models—Artificial Neural Networks (ANN), k-Nearest Neighbours (KNN), Support Vector Machine (SVM), Naive Bayes (NB), and Linear Regression (LR)—have been developed to predict patient health outcomes based on sensor data. Performance metrics and confusion matrices demonstrate the remarkable abilities of these models, with ANN standing out as a top performer. By combining IoT and machine intelligence, healthcare professionals can manage their patients’ treatment proactive and intervene early. This study highlights the revolutionary potential of machine learning and the internet of things to improve patient outcomes, monitor kidney health more effectively, and cut healthcare costs. As healthcare systems develop, the use of IoT and machine learning to manage diseases will revolutionise patient care.

Cough serves as a principal symptom in respiratory conditions. Variations in cough sound characteristics provide valuable diagnostic insights. There is a lack of evidence in characterizing cough sounds and misinterpretation leads to severe consequences. This paper presents the efficient characterization of cough sounds using statistical analysis; in addition, both cough sound and speech characteristics are compared. The proposed method extracts spectral and time domain attributes, further subjected to statistical and histogram analysis. The results show that the 25th percentile of spectral roll-off and spectral flux, along with the maximum and mean values of spectral flatness, are vital for characterizing cough sounds. Additionally, the maximum and 75th percentile of zero crossing rate, median of spectral bandwidth, and minimum, maximum, median, mean, standard deviation, 25th percentile, and 75th percentile of spectral centroid contribute significantly to this characterization. The distribution of features in cough sounds discloses that spectral roll-off spreads up to 7800 Hz, spectral flatness ranges from 0 to 0.22, spectral flux varies between 0.3 and 0.6, zero crossing rate extends up to 0.4, spectral centroid spans up to 4300 Hz, and spectral bandwidth varies between 1300 Hz to 3200 Hz. Using these attributes as inputs for artificial intelligence models thereby improves respiratory disease diagnosis efficiency.

The primary objective of this research paper is to develop an efficient method for the early identification of heart failure. Two classification techniques—Logistic Regression (LR) and Naive Bayes (NB)—were used in a series of experiments utilizing the heart failure dataset from the UCI repository. The authors selected accuracy as the performance metric and applied the robust feature selection technique to both classifiers. The experimental approach systematically excluded each prediction attribute and calculated accuracy with the remaining features. Notably, when the Platelets feature was removed, both LR and NB obtained a remarkable 100% accuracy. This significant finding highlights the potential of the suggested model for the early detection of heart failure. This research gives priceless insights that might empower doctors in improving heart failure diagnosis and patient care by identifying critical predictors. The proposed model shows potential for medical practitioners engaged in diagnosing heart failure.

The COVID-19 pandemic has created a huge challenge for healthcare services around the world. Understanding the factors affecting treatment outcomes in COVID-19 is important to provide personalized and effective treatment, especially taking into account gender differences. This challenge involves using machine learning to analyze patient data, identify risk factors, and develop predictive models to predict the incidence and severity of COVID-19, including the impact of gender on the disease. This will allow doctors to create treatment plans and allocate resources efficiently based on a person’s gender and other health-related factors. The aim of this article is to develop and evaluate novel machine learning algorithms to predict the clinical outcome of COVID-19 in patients, including the effect of father’s gender. The goal is to develop accurate predictive models that will help doctors predict the progression and severity of COVID-19 in humans, including gender-specific factors in our study I used 7 different ML model with different k fold cross validation and compare with other model to our proposed model and got 99.45 accuracy and other model like random forest accuracy is 94.34, SVM has 95.88 accuracy and other model has got accuracy bellow 95 so we conclude that our proposed model has got best accuracy.

One of the most critical neurological disorders is a brain tumour, characterized by the uncontrolled proliferation of abnormal cells within the brain. The incorporation of cutting-edge automated technology is crucial to enhance the accurancy of tumour detection. Glioma, meningioma, pituitary, and normal brain are the four groups targeted for classification in MRI scans of the brain. Convolution neural networks that have been extensively trained, including AlexNet and VGG19, are frequently utilized for image categorization utilizing transfer learning. However, due to the significant storage space needs, they cannot be used successfully on edge devices to build robotic devices. Therefore, the classification procedure was carried out using a genetic algorithm, which takes up around 30–40% less space than the original model and reduces inference time by about 50%. Before compression, the accuracy given by AlexNet and VGG19 was 86.12% and 94.78%, respectively, and the accuracy after compression for AlexNet and VGG19 was 87.12% and 92.04%, respectively.

Using ensemble learning toward medical diagnostics as a response to diabetes on a global scale. The data set is composed of medical and demographic information collected from survey questionnaire forms filled out by patients; medical charts; and lab samples from diagnosed or at-risk subjects during patient clinic visits and hospitalizations. For instance, variables include age, sex, obesity, hypertension, ischemic heart disease, prior smoking status, post-prandial test blood, and random blood sugar levels in non-diabetic subjects. Rigorous data processing ensures reliability. Ensemble learning emphasizes the potential of predicting diabetes, thus, giving a more accurate forecast on the same and advanced prevention techniques. The approach is also useful in detailed research on the origins of diabetes and providing guidelines for prevention and treatment campaigns worldwide. The study reveals a highly accurate classification model with an overall accuracy of 95%. Precision is notable, with 95% for class 0 and 93% for class 1, while class 0 exhibits outstanding recall at 99%, whereas class 1 has a lower recall at 61%.

Breast cancer is a significant contributor to the death rate of women in developing and underdeveloped nations. Timely identification and categorization of breast cancer can facilitate the administration of the most optimal therapy to patients. Using ensemble learning, we presented a novel deep-learning architecture for breast cancer detection and classification in breast ultrasound images. In the proposed work, image features are extracted using three pre-trained CNN architectures, DenseNet121, DenseNet169, and DenseNet201, which are then averaged to form an ensemble model. Experiments are conducted using Kaggle’s publicly available data set to evaluate the performance of the proposed architecture. Regarding accuracy in detecting and classifying breast cancer in ultrasound images, it has been visible that the proposed ensemble architecture outperforms other pre-defined deep learning architectures with an accuracy of 99.62%.

Breast cancer is a major contributor to cancer-related death in women. A higher likelihood of survival could result from early detection if the patient could receive the appropriate medicine while it is still in its early stages. Most often, a medical professional will use medical imaging or manual physical analysis to make a diagnosis. These efforts might be drastically cut with an automated approach. Using deep learning approaches, this paper proposes a system for autonomously analyzing ultrasound pictures. Using data obtained from the web repository Kaggle, three deep learning models—InceptionV3, VGG16, and VGG19—are applied to validate the suggested method. With the help of a confusion matrix and accuracy metrics, we compare the outcomes produced by these three deep learning methods. With an accuracy rate of 99.75%, the InceptionV3 model proved to be the most effective.

Human society has confronted numerous threats throughout its history, ranging from the Bubonic Plague to the recent COVID-19 pandemic. Despite the challenges, solutions have been developed for most issues, often at an affordable cost. However, one persistently expensive and life-threatening problem remains: cancer. Cancer acts like a societal curse, impacting individuals on mental, physical, and financial fronts. While remedies exist, they come at a significant cost, and survival rates, particularly for lung cancer, are often discouraging. Early detection is crucial, but human error in interpreting CT scans and MRI scans can lead to catastrophic consequences. To address accuracy concerns, an AI-based approach has been devised, aiming to minimize errors. This innovation significantly increases the likelihood of early problem detection, enabling prompt initiation of life-saving treatments. The current tested algorithms show promising results, empowering medical professionals, especially oncologists, to identify lung cancer in its early stages. This advancement holds the potential to substantially improve patients’ chances of survival.

CRC or Colorectal Cancer also called as bowel cancer is the development of cancer from colon or rectum. Detection of CRC can be very essential that can help the diagnosed with effectual treatment preventing potential loss of life. Conventional methods can be challenging because of its excessive dependence on the expert to detect accurately. This paper aims to compare results obtained from popular deep learning models such as AlexNet, GoogleNet, MobileNet, by performing on the “CRCHistoPhenotypes” dataset furthermore, inter-comparison of the same is done by applying Data Augmentation methods. Comparison is done on the basis of training time, accuracy, weighted f1 score, specificity and sensitivity. An enhancement in testing accuracy was observed, even in the case of the state-of-the-art network, GoogLeNet. It exhibited an increase of around 2.3%, achieving an impressive 80% accuracy following the utilization of data augmentation methods.

IoT devices are widely used in medical domain for detection of high blood sugar and life threatening disease such as cancer. Breast cancer is one of the most challenging type of cancer which not only affects women but in some cases men also. Deep learning is one of the widely used technology which provides efficient classification of cancerous lumps but it is not useful for IoT devices as the devices lack resources such as storage and computation. For the suitability in IoT devices, in this work, we are compressing UNet, the popular semantic segmentation technique, for the pixel-wise classification of breast cancer. For compressing the deep learning model, we use genetic algorithm which removes the unwanted layers and hidden units in the existing UNet model. We have evaluated the proposed model and compared with the existing model(s) and found that the proposed compression technique suppresses the storage requirement to 77.1%. Additionally, it also improves the inference time by 3.82 $$\times $$ × without compromising the accuracy. We conclude that the primary reason of inference time improvement is the requirement of less number of weight and bias by the proposed model.

In this study, we propose a cost-effective computer-aided detection system based on machine learning for the classification of colorectal cancer tissues. Colorectal cancer stands as the third most prevalent cancer globally and is the second leading cause of malignancy-related deaths. The proposed computer-aided detection system involves partitioning each image out of 7180 histopathological images into 16 equal-sized blocks. Subsequently, features are extracted from each block of image in RGB (red, green, blue), HSV (hue, saturation, value), and L*a*b* color spaces. The extracted features include regional and gray-level co-occurrence matrix features. Following the extraction, these features undergo scaling to eliminate outliers before being input into a machine learning classifier. The performance of the machine learning models is enhanced by optimizing the hyperparameters of the models. Notably, CatBoost outperformed all other models, achieving an exceptional accuracy of 95.19%. This remarkable accuracy indicates CatBoost as a promising model for the task of colorectal cancer tissue classification.

The prognosis and survival chances for people with breast cancer can be significantly improved by an early diagnosis. Therefore, it is crucial to accurately identify malignant tumors nowadays, it has become a frequent health problem and its occurrence is also increased and has high morality. It is also increased due to unawareness and change in the lifestyle of women. It is quite difficult to detect it in the early stage. It is also the deadliest disease after lung cancer. The most optimal machine learning technique to utilize to diagnose a certain disease is still a debate because different things can affect how accurate the results are. Hence, it is mandatory to devote effort in building up a strategy that produces fewer mistakes while improving precision. The research compares four algorithms SVM, Logistic Regression, Random Forest, and KNN—that prognosis the course of breast cancer using various datasets. Following a precise comparison of our models, we discovered that KNN outperformed all other algorithms and had a better efficiency of 97.8%. And, KNN has proven to be effective in predicting and diagnosing breast cancer and gives the best results in terms of accuracy and precision. Improved accuracy by using a variety of algorithms on the basis of the data set and model’s predictions also did a fantastic statistical analysis.