Modern project management demands the mitigation of risk factors in conjunction with optimal resource utilization. Traditional approaches to risk mitigation, based on subjective assessment and supportive static tools, cannot operate effectively in a project environment filled with dynamics. Advanced analytics, artificial intelligence, and deep learning drive data-based decision-making and are able to bring forth predictive insights leading to adaptive management strategies. This paper proposed advanced deep learning approaches to improve risk reduction and optimization of resource allocation in software project management systems. First, the raw historical dataset, collected from the sources that are freely available, is gathered and preprocessed. The dataset collected contains a lot of irrelevance, which causes prediction error and computation complexity. In this regard, the pre-processing, such as one-hot encoding and normalization of data, is done, which reduces issues of data redundancy. After pre-processing, a stacked bidirectional long short-term network S-BiLSTM is proposed for the joint prediction of risks and enhancement of decision power for resource scheduling. Further, optimal resource allocation is carried out by the employment of the Draco lizard optimization DLO technique based on different parameters comprising project complexity, integration complexity, schedule pressure, team size, team experience, technology risk, and risk management maturity. The proposed framework is analyzed through a simulation on the Python platform. Various factors, such as accuracy, workload distribution analysis, resource optimization efficiency , convergence analysis, training time, and resource utilization rate, are analyzed and compared with conventional approaches to demonstrate the proposed approach’s efficiency. The proposed approach achieves a predictive accuracy of 99.2%, significantly improving the performance of the PMS in project risk identification.

Alzheimer's disease (AD) is one of the most common forms of dementia worldwide. As the condition worsens, it makes it harder for a person to do daily chores independently. The most common illness, Alzheimer's, begins slowly with modest cognitive impairment and may eventually lead to loss of environmental awareness and communication. For clinical intervention to be successful, AD must be diagnosed accurately and promptly. Even seasoned radiologists may find it difficult to interpret AD from medical imaging. In order to effectively and consistently identify AD traits from Magnetic Resonance Imaging (MRI) images, this study proposes a framework called ViT-BiGRU, which combines a Vision Transformer (ViT) with a Bidirectional Gated Recurrent Unit (BiGRU). Initially, the input MRI is pre-processed using image rescanning, and filtering is done using an Adaptive Gaussian Filter. The Red, Green, and Blue (RGB) Grayscale produces accurate classification results. The pre-processed image is subjected to ViT-BiGRU for classifying output as AD, Cognitively Normal (CN), Early MCI (EMCI), and Late MCI (LMCI). Hyperparameters that tune the ViT-BiGRU model use Guided Remora Optimization (GRO) to increase the training speed. The ADNI dataset is also used in this method. The performance metrics are accuracy, precision, recall, specificity, kappa, f1 score, RMSE, MSE, MAE, and MCC, which attain high performance. The accuracy of the proposed method is 99.45% compared to the existing model, and the error rate is low compared to the existing method. Also, the proposed approach provided insights into disease identification by utilizing Grad-CAM and Grad-CAM++ to provide visual explanations.

This work presents an Attention-Based Fuzzy BiLSTM (A-FBiLSTM) model that builds upon our previous research utilising BiLSTM and Fuzzy Logic for the identification of hand motions via EEG signals. The suggested model has LSTM encoders that work in both directions, a temporal attention mechanism, and a fuzzy inference decision layer. These properties help it identify things apart, make it more dependable, and make it easier to understand when EEG data is unclear or when there are variances across participants. We assess the method utilising publicly accessible EEG motor imaging datasets, including PhysioNet EEGMMIDB, and contrast it with baseline models such as CNN, BiLSTM, and BiLSTM + Fuzzy. The findings of the experiment demonstrate that A-FBiLSTM has an accuracy of 97.12%, an F1-score of 0.96, and an AUC of 0.99. The earlier BiLSTM + Fuzzy technique was 1.8–2% less accurate than the present one. Also, visualising attention weights demonstrates how essential EEG features are in time and space, which shows that the model works. These results suggest that A-FBiLSTM is a very accurate and easy-to-understand technique to employ BCI in real time, like for controlling prosthetics and neurorehabilitation.

This study addresses the global challenge of detecting and treating psychiatric conditions like depression. Traditional methods for assessing depression, such as patient interviews and PHQ-9 scores, are often limited by their subjectivity and lack of granularity. To overcome these challenges, this research introduces hybrid deep learning (DL) architectures that that leverages the complementary strengths of BERT for contextual language understanding and sequential models for analyzing temporal patterns in speech or text. Using the DAIC-Woz dataset, our study explores both text and audio modalities. We first implement baseline non-BERT models (Text-CNN, Audio-CNN, GRU, and BiLSTM) and then develop hybrid BERT-based models (BERT-CNN, BERT-Audio-CNN, BERT-GRU, and BERT-BiLSTM) to capture both contextual semantics and sequential patterns. Experimental results demonstrate that BERT-enhanced models consistently outperform non-BERT baselines, with the BERT-BiLSTM achieving the best performance (93.6% accuracy). Overall, the proposed hybrid approach advances state-of-the-art (SOAT) multimodal depression detection by improving accuracy, robustness to data imbalance, and generalization across modalities.

A vital input for a task using the brain in BCI (Brain Computer Interface) applications is the motor imagery (MI) signal from the brain. Users of BCI systems can operate external equipment by using their brain activity, using motor imagery as a control method. Innumerous Electroencephalography (EEG) channels are used to gather nerve impulses from the brain, which are the most prevalent input for Brain Computer Interface systems while they are minimally invasive, flexible, and low in price. The computational overhead is increased by multichannel BCI systems’ high-dimensional data, which causes processing to be slower and to cost more money. EEG details are regularly gathered from over 100 different brain regions; therefore, it is essential to use channel selection algorithms to choose the ideal channels for a given circumstance. However, the primary objective of choosing the channel in EEG data analysis is to lessen the computer intricacy, improve the precision of classification by eliminating over fitting, and save setup time. In this study, we suggested a remora optimization technique that was inspired by nature to lessen the computational load brought on by several channels. Using predetermined criteria, a number of channel selection evaluation techniques, whether classification-based methods used or not it extracted the proper channel subsets. In order to determine the greatest classification accuracy, the classification procedures were utilized in the end. Three publicly available EEG datasets are used to validate the experiment (BCI Competition IV-1,2a, Competition III-3a), and it resulted in superior classification accuracy.

Nowadays, skin cancer became a common disease where every 3 in 100 people are affecting from skin cancer. Previously doctors used to detect the cancer efficiently, but nowadays they are unable to detect. So, there is a drastic demand for computer-based detection systems. Usually, computer detects skin cancer from dermoscopic images by using deep learning techniques. Many researchers used ML and DL techniques to detect skin cancer and to find accurate results. But they did not perform well for new images. In this paper, we added CNN model to predict the skin cancer along with that, we proposed a model called Meta Block which has metadata along with dermoscopy images of patients that includes all records of patients that helps in predicting the cancer more accurately. In this paper, we used two different datasets for skin cancer classification and we used 5 different models and compared the results with previous research results. We found that by comparing the accuracies are increased by 10

Hand-motion recognition using EEG data is one potential method for developing non-invasive brain-computer interfaces (BCIs) that can assist individuals with motor impairments. This work uses Bidirectional Long Short-Term Memory (Bi LSTM) networks to present a novel approach for identifying hand movements based on EEG data. The EEG data collected from scalp electrodes captures the brain activity associated with motor goals. We preprocess the raw EEG data to find meaningful features correlating to unique hand movements. These properties are then used as input sequences by the Bi LSTM network, which can learn temporal dependencies both forward and backward. We evaluate our method using publicly available EEG datasets from hand gesture experiments including various subjects. We conduct extensive experiments and compare our Bi LSTM-based method with baseline techniques to demonstrate its effectiveness in accurately recognizing different hand movements from EEG data. In our project, we divided the data set into 80 percent and 20 percent sections for the training and testing stages, respectively, and achieved an accuracy of 95.34 percent. Our findings demonstrate how Bi LSTM networks can interpret complex brain signals in real time for uses such as prosthesis control and rehabilitation.

One of the most important aspects of classification is choosing features in such a way as to get rid of redundant or irrelevant elements in the dataset. For the most part, multi-objective feature selection strategies have been offered by a number of scholars as a strategy for this aim. On the other hand, these techniques frequently fail to simultaneously improve classification accuracy while removing redundant feature combinations. This article presents a wrapper-based feature selection strategy that strikes a compromise between classification accuracy and redundancy reduction by combining features of the multi objective (MO) based honey badger algorithm (MO-HBA) and non-dominated sorting genetic algorithm-II (NSGA-II). The technique was developed as part of this investigation. Increasing the accuracy of the classification while simultaneously reducing the number of redundant characteristics is one of the optimizations aims of this approach. The MO-HBA shows excellent performance in exploration and exploitation. A Kernel version of the extreme learning machine (KELM) is used for the process of selecting the features to use. In order to evaluate how well this method of feature selection performs, eighteen benchmark datasets are utilized, and the results are compared to four established methods of multi-objective feature selection based on different metrics.

This paper aims to investigate the efficacy of EEG-based stress detection using a Random Forest classifier during the Stroop Test, a key psychological assessment probing cognitive functions such as processing speed, cognitive flexibility, and attentional capacity. The study involved 10 participants (5 male,5 female), aged 18 to 30, all with normal or corrected-to-normal vision and no documented cognitive impairments. EEG data was recorded using PsychoPy software and standardized Stroop Test stimuli. Signal processing, including notch and bandpass filtering, was applied to eliminate noise and emphasize relevant brainwave frequencies. Artifact rejection, notably Independent Component Analysis (ICA), was utilized to remove physiological artifacts, and epoching was used to segment the EEG data for detailed analysis. A Random Forest classifier, trained on the preprocessed EEG data, achieved an accuracy of 83.33% in distinguishing between different cognitive states. These findings highlight the efficacy of combining signal processing and machine learning techniques for cognitive state detection through EEG analysis.

Electroencephalography (EEG) has great potential for deciphering the motor activity of the human brain, especially when imagining an individual performing a specific motor task. This ability to interpret EEG signals and determine imaginary motor activity has important implications for the development of brain-computer interfaces (BCI) and advances in neurorehabilitation techniques. In this work, we investigate the application of deep neural networks, particularly convolutional neural networks (CNNs), to analyze EEG signals [2] in an image-like format and classify the corresponding motor activity. The dataset consists of EEG recordings collected from subjects performing various motor tasks or motor imagery. We use preprocessing techniques to remove noise and artifacts and then segment the EEG signal into smaller epochs. These epochs are transformed into an image-like representation suitable for CNN input. Data augmentation is applied to increase the size and diversity of the data set, ensuring robust network training. CNN architectures are designed to capture spatial patterns and hierarchical features present in EEG signals. The network is trained using a suitable loss function and optimization algorithm. Performance evaluation is performed against a separate validation set using metrics such as accuracy, precision, recall, and F1 score [4]. This result demonstrates the effectiveness of the CNN-based approach in interpreting EEG signals and classifying imaginary motor activity. The results will contribute to the development of more accurate and efficient BCIs, advances in neurorehabilitation, and advances in our understanding of motor representations in the brain.

Data availability is one of the primary principles of information security. Distributed Denial of Service (DDoS) is a typical cyber security attack of the DOS family to deny data availability to legitimate users by flooding the system resources with huge amounts of network traffic. This study is a quantitative research to evaluate SVM kernels for Reflection Amplification DDoS attack detection with different uncorrelated feature subsets. Support Vector Machine (SVM) classification algorithms Linear, RBF, Poly, and Sigmoid kernel functions are used in this study. This research proposed three methods to select uncorrelated subsets for SVM to classify DDoS attacks. This study performs experiments on Reflection Amplification DDoS attacks of SNMP and DNS datasets. The first proposal, selects uncorrelated feature subsets by Pearson, Spearman, and Kendall correlation methods. In the second proposal, common uncorrelated feature subsets of three correlation methods are called PSK-uncorrelated feature subsets. The third proposal, common PSK-uncorrelated feature subsets of SNMP and DNS datasets are called Reflection_DDoS uncorrelated feature subsets. SNMP and DNS data sets are collected from the CIC-DDoS2019 evaluation datasets. Experiment results conclude that Poly and RBF kernels among the SVM kernel functions produce the best classification results and among uncorrelated feature subsets, Pearson uncorrelated feature subset produces the best results in three un-correlated feature subsets.

An important task for classification is feature selection that removes the redundant or irrelevant features from the dataset. Multi-objective feature selection approach is mainly proposed by many researchers. However, these approaches failed to maintain the higher classification accuracy while removing redundancy in the features. In this work, a wrapper based feature selection technique is proposed with a hybrid of Multi Objective Honey Badger Algorithm (MO-HBA) and Strength Pareto Evolutionary Algorithm-II to maintain the balance between classification accuracy and removal of redundancy. Classification accuracy improvement and removal of redundant features are considered as the multi-objective optimization functions of the proposed multi-objective feature selection technique. The Levy flight algorithm is utilized to initialize the population to enhance the ability of the exploration and exploitation of MO-HBA. The regularized Extreme Learning Machine is used to classify the selected features. To evaluate the performance of the proposed feature selection technique, eighteen benchmark datasets are utilized and results are compared with the four well known multi-objective feature selection techniques in terms of accuracy, hamming loss, ranking loss, mean value, standard deviation, length of features, and training time. The proposed approach achieved maximum accuracy of 100% with the maximum value of selected features as 80. The minimum value of hamming loss, ranking loss, mean value and standard deviation value achieved by the proposed approach are 0.0092, 0.0003, 0.018 and 0.001 respectively. The experimental results show that the proposed approach can give improved classification accuracy while the removal of redundancy in large scale datasets.

In solving two or more objective problems, multi-objective evolutionary algorithms (MOEAs) have proven their effective performance. In most of the MOEAs based feature selection algorithms, more optimal solutions are obtained around the Pareto front's center because of the deficiency in selection features. With the complicated Pareto fronts, it is difficult to select features using the penalty boundary intersection decomposition approach that can provide stable selection pressures. With the adaptive penalty boundary intersection (APBI) decomposition approach, a novel wrapper based feature selection technique named MOGHBNS3/D is proposed in this paper based on the hybrid of Multi Objective Guided Honey Badger Algorithm and Non-dominated sorting genetic algorithm III. Classification accuracy improvement and removal of redundant features are considered to be the multi-objective optimization functions of the proposed multi-objective feature selection technique. An external archive is employed as a repository for storing the non-dominated solutions in the MOGHBNS3/D. To enhance the selection pressures of the external archive, the penalty values are adjusted adaptively using the APBI mechanism. The Wavelet kernel Extreme Learning Machine is used to classify the selected features. To analyze the performance of the proposed feature selection technique, eighteen benchmark datasets are utilized and results are compared with the four well known multi-objective techniques in terms of accuracy, precision, recall, F1 score, coverage error, hamming loss, ranking loss and training time. The experimental results show that the proposed method can select optimal features with improved classification accuracy, diversity and convergence while the removal of redundancy.

Lung cancer is a dangerous tumor that requires accurate diagnosis for effective treatment. Traditional diagnosis involves invasive and time-consuming histologic examination, and radiologists face challenges in localizing lung tumors. Deep neural convolutional networks are frequently used to locate lung cancer, but this is still difficult when not accounting for surrounding lung tissue. Despite progress in research, healthcare still uses deep learning models to improve the precision and sensitivity of large datasets. CNN (Convolutional Neural Network) accuracy standards are adequate, but image properties such as flips, construction, and other uncommon alignments diminish its efficiency. CNN also does not store the geometric distribution between scanned picture features. CT (Computed Tomography) and PET (Positron Emission Tomography) scans require a method that takes into consideration the spatial information of picture characteristics, as they are vulnerable to alignment problems during the perusing process. To address these issues, the authors propose MCNet (MobileNetV2 with Capsule Network), a hybrid network that adopts feature extraction and categorization from MobileNetV2, and capsule network is used to overcome the limitations of convolutional neural networks (CNNs) when it comes to processing images with abnormal orientations, such as tilting or rotation. Although CNNs are effective in processing images presented in a standard orientation, they have difficulty handling variations in image orientation. In this work, MobileNetV2 serves as a backbone network for Capsule Networks in lung cancer diagnosis. The lung image collection dataset verifies the effectiveness of MCNet, and experimental results show that MCNet technology performs better than previous state-of-the-art techniques. The proposed hybrid MCNet architecture achieves the clinical goal of lung cancer diagnosis with a lower computational cost, reducing processing time complexity and false positive rates compared to current techniques.

In this analysis, Optimization of Hybrid Grey Wolf technique for Feature Selection analysis is implemented. In many areas standard performance is given by Grey Wolf Optimizer's (GWO). To enhance the performance different GWO factors are signified GWO mainly based on the two conflicting concepts. This will exploit the optimal solutions. The performance of search algorithms is improved by balancing the both exploration and exploitation. For continuous search space problems are obtained from the original hybrid approach which is superior to performance. The main binary problem is fromfeature selection. Hence hybrid PSOGWO is also known as BGWOPSO. This is the best feature subset. To produce the best results, Euclidean separation matrix and K-Nearest Neighbor's classifier are used. Performance is evaluated using one common benchmark dataset from the UCI repository. According to the findings, the whale optimization algorithm, binary genetic algorithm, binary PSO, binary genetic algorithm, and Binary GWO (BGWO) provide the highest levels of performance in terms of accuracy, the selection of the most effective optimal features, and the amount of time required to compute.

The study addresses the challenges of human action recognition and analysis in computer vision, with a focus on classifying Indian dance forms. The complexity of these dance styles, including variations in body postures and hand gestures, makes classification difficult. Deep learning models require large datasets for good performance, so standard data augmentation techniques are used to increase model generalizability. The study proposes the Indian Classical Dance Generative Adversarial Network (ICD-GAN) for augmentation and the quantum-based Convolutional Neural Network (QCNN) for classification. The research consists of three phases: traditional augmentation, GAN-based augmentation, and a combination of both. The proposed QCNN is introduced to reduce computational time. Different GAN variants DC-GAN, CGAN, MFCGAN are employed for augmentation, while transfer learning-based CNN models VGG-16, VGG-19, MobileNet-v2, ResNet-50, and new QCNN are implemented for classification. The study demonstrates that GAN-based augmentation outperforms traditional methods, and QCNN reduces computational complexity while improving prediction accuracy. The proposed method achieves a precision rate of 98.7% as validated through qualitative and quantitative analysis. It provides a more effective and efficient approach compared to existing methods for Indian dance form classification.

In a human life there are alot of life-threatening consequences, one among those dangerous situations is having a brain stroke. When part of the brain does not receive sufficient blood flow for functioning a brain stroke strikes a person. This is most often due to a blockage in an artery or bleeding in the brain. When brain cells are deprived of oxygen for an extended period of time, they die. If enough brain cells in this area die, the damage becomes irreversible and you may lose the ability you once had control over. However, restoring blood flow can prevent or at least limit the severity of this type of damage. Therefore, stroke treatment is time-critical. Different areas of the brain control different abilities hence the symptoms for a brain stroke are not constant for everyone. The brain stroke is what a heart attack does to the heart is what a stroke does to the brain. When a stroke occurs, part of the brain loses its blood supply, leaving that area of the brain without oxygen. Without oxygen, the affected brain cells are starved of oxygen and stop functioning normally. Strokes are very common. Stroke is the second leading cause of death worldwide. This paper is based on the prediction of brain stroke using machine learning algorithms which helps to rehabilitate the patient so that one can gain their life back to normal.

Automatic detection and segmentation of liver tumors from Computed Tomography (CT) scans is a challenging task due to their complex morphology in CT images, but advances in deep learning have made it possible to achieve high accuracy in the segmentation and detection of liver tumors, eliminating the need for manual intervention and allowing radiologists to focus more on critical diseases by giving awareness about the liver tumor at an early stage, which may save a person’s life by allowing them to receive treatment sooner, possibly preventing further severity or death. In our study, we worked on different approaches to detect liver tumors from CT scans and analyzed their performance in terms of accuracy. A few of the approaches employed are using Convolutional neural networks including existing ones like U-Net, ResNet50, while we proposed a combination of multiple Convolutional Neural Network models (CNN) to obtain better results. It was observed that the combination of CNN models produced higher accuracies rather than the existing models.

During the pandemic, online transactions have become a usual activity for any purchase. Use of credit card has been growing rapidly, as a result, the number of frauds regarding the credits cards has increased enormously. This includes fraudulent transactions or transfer of money from our accounts. In our paper, an evaluation of machine learning, artificial intelligence, and deep learning algorithms like CNN, Local Outlier Factor, Decision Tree, Cat Boost, XG Boost, Isolation Forest, ANN, K-Means, Naive Bayes, Ada Boosting, Bagging, Multi-Layer Perceptron, KNN, Logistic Regression, Random Forest, SVM, Gradient Boosting, Over Sampling, SMOTE, Extra tree, have been contemplated for credit card fraud detection. In the previous papers where time parameter is considered during the credit card fraud detection whereas, in our paper, we will be considering other factors like location as well as the amount for transaction done. As a result, a distinction was made among different algorithms and datasets based on the factor accuracy.

In machine learning, a crucial task is feature selection in that the computational cost will be increased exponentially with increases in problem complexity. To reduce the dimensionality of medical datasets and reduce the computational cost, multi-objective optimization approaches are mainly utilized by researchers. Similarly, for improving the population diversity of the Flamingo Search Algorithm, the neighbourhood centroid opposition-based learning mutation is employed. In this paper, to improve the classification accuracy, enhance their exploration capability in the search space and reduce the computational cost while increasing the size of dataset, neighbourhood centroid opposition-based learning (NCOBL) is integrated into the multi-objective optimization based Flamingo Search Algorithm (MOFSA). The optimal selected datasets are classified by using the weighted K-Nearest Neighbour classifier. With the use of fifteen benchmark medical datasets, the efficacy of the suggested strategy is assessed in terms of recall, precision, accuracy, running time, F-measure, hamming loss, ranking loss, standard deviation, mean value error, and size of the selected features. Then the performance of the suggested feature selection technique is compared to that of the existing approaches. The suggested method produced a minimum mean value, standard deviation, mean hamming loss, and maximum accuracy of about 99%. The experimental findings demonstrate that the suggested method may enhance classification accuracy and also eliminate redundancy in huge datasets.