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.

In the rapidly evolving digital landscape, trend forecasting has become a critical task for decision-makers across industries. Traditional methods struggle with adaptability, scalability, and real-time trend identification. This paper presents a novel framework that integrates Generative AI with the Proposed Guided Remora Optimization Algorithm (PGROA) to enhance trend prediction accuracy while maintaining robustness across dynamic and multimodal datasets. The framework leverages transformer-based architectures for feature extraction, adaptive learning mechanisms for real-time updates, and cross-domain generalization techniques to ensure scalability. Additionally, interpretability methods such as SHAP values and attention mechanisms provide transparency in model predictions. The proposed system is evaluated on diverse datasets, demonstrating superior performance with an accuracy of 94.8%, an F1-score of 93.8%, and a significantly reduced RMSE of 0.072, outperforming existing deep learning and hybrid models. This research establishes a scalable and interpretable AI-driven approach to trend prediction, equipping decision-makers with actionable insights for dynamic environments.

Diabetic retinopathy is a condition resulting from harm to the veins and blood vessels of the retina. It might begin with little or no signs and progress to impaired or possibly absent vision. It is critical to get regular vision tests for diagnosis. Regulation of insulin levels and, in critical circumstances, the nerve that connects the eyes is placed in hazard by glaucoma that often occurs along with high levels of intraocular pressure. This may contribute to complaints including vision loss in the direction of vision. These medical conditions underscore the significance it is to undergo regular vision examinations in order to preserve eye health and promptly recognize and address any issues that arise. Skilled professionals must identify and interpret numerous minor anomalies. This study employs ResNetV3, VGG16, and CNN architectures to provide a unified deep neural networks strategy for DR identification. Current DR detection tools (Bogdănici et al. in Rom J Ophthalmol 62:112, 2017) [1] rely heavily on ophthalmologists for manual evaluation. To solve this issue, we developed a ResNetV3 network that diagnoses DR through virtual retinal images. Optimized ResNetV3 is designed to detect certain traits including color vision impairment, diabetic retinal detachment syndrome, glaucoma, and visual difficulties. This computerized system aims to increase diagnosis accuracy by offering rapid and effective answers with minimal human participation.

Since many years smartphones are utilised for human activity recognition (HAR), important healthcare recommendations and telemedicine. Deep learning (DL) and machine learning techniques are commonly employed in studies of statistical models of human behaviour. However, the performance of present HAR platforms is constrained by complex physical activities. In this study, we developed the Ada-HAR human activity identification and real-time monitoring system, which is able to recognise more human motions in erratic situations. The chosen hierarchical clustering and classification methods are able automatically identify and categorise 12 activities (five dynamics, six statics, and a series of transitions). Finally, actual tests were run to verify the effectiveness and reliability of the suggested methods. The results demonstrate that the DL-based classifier achieves a higher identification rate (95.15% for waist and 92.20% for pocket) in comparison to the techniques discussed in the literature. Finally, the Ada-HAR system can track human behaviour in real-time regardless of how the smartphone is pointed. Here sensor activity dataset is considered which consists of user's activity log. By applying algorithms named LSTM, Adamax, Adagrad, SGD, indentifiesExi user’s activity based on the movement of the mobile. Existing Algorithms named LSTM, K-NN, DT, ANN, SVM, NB applied on different dataset produced better result and accuracy. RNN is an updated version of LSTM. RNN stores only the current activities and fails to store previous conclusions. Fortunately, LSTM stores existing and completed work in any dynamic situations. The LSTM works with random weights which led to local optimum and high time complexity. This paper has addressed the above issue by the usage of optimization algorithms in weight updation of LSTM. This increase the accuracy of the model. Here Adamax, Adagram, Stochastic gradient descent are used for weight updation.

The diabetic retinopathy (DR) is the leading cause of blindness and occurs when the tiny blood vessels in the retina are damaged. Since DR is a silent disease that may not cause any signs or only cause mild vision problems, it is important to get an eye exam every year so that it can be found early and treated more effectively. Fundus cameras are used to take images of the retina during an eye exam. But for a number of reasons, there is a chance that the images will be blurry and not good enough for a right diagnosis. Because of this, there is a need to improve low-quality images with the right tools. Contrast limited adaptive histogram equalization (CLAHE) is a famous way to improve the quality of a retinal image. In this work, an improved Wiener filter (IWF) is used with the Enhanced CLAHE (ECLAHE) enhancement method to improve the quality of retinal images even more. The IWF can change itself on a local level by tuning its kernel to keep edges and features while reducing noise effectively. Fundus images also have another problem, which is that the lighting isn’t always even. In this study, a method called gamma correction (GC) was used to avoid these kinds of problems. The Local Digital Diabetic Retinopathy (LDDR) database, which is a collection of retinal images, was used to test the benefits of image enhancement. The results were compared with standard CLAHE, Weiner Filter, Gamma Correction, and combination ways of retinal enhancement. Experiments showed that our hybrid method results that were on par with those of the other enhancement methods.

One of the more prevalent psychological disorders is depression, and numerous grief individuals contemplate suicide each year. Because people feel embarrassed or inexperienced whether they’re suffering from depressive symptoms, patients with depression generally skip asking for the guidance of licensed mental health professionals, which might lead to an enormous gap in receiving appropriate treatment. In the meantime, research suggests that online social networking data offers helpful information regarding physical and psychological issues. Throughout this paper, we claim that by analyzing online social behaviors, depression might be detected early on. To achieve accurate depression diagnosis, the machine learning technique SVM is utilized, which is innovative in the area of depressive disorder identification. This does not depend on the extraction of numerous or multifaceted characteristics. Once it regards depression identification, algorithms based on machine learning provide several significant benefits over traditional methods of statistical analysis. Basic continuous relationships were rare in depression symptoms. Highly accurate forecasts can be generated by machine learning algorithms as they are capable of learning effectively from complex, irregular structures in data. Despite human feature extraction, which is a time-consuming and laborious action in traditional approaches, these techniques are capable of autonomously retrieving relevant characteristics from huge data sets. Utilized these five algorithms for predicting depression through social media: Support Vector Machine, Artificial Neural Network, Deep Neural Network, CatBoost, and Long Short-Term Memory. Perhaps the most efficient of these methods is the Support Vector Machine.

Many applications require effective classification of imbalanced data, which is found everywhere. Existing classification algorithms often misclassify the minority class in imbalanced data due to the dominant class’s influence. Boosting algorithms combine basic learners to improve their performance. AdaBoost, a popular ensemble learning system, can classify general datasets well. But this algorithm will be limited misclassified samples only. The minority-classified samples are not fit for this algorithm and as it alone not readies for imbalanced data classification. This paper introduced multi-level strategy to solve imbalanced data, where combined SMOTE with AdaBoost to process unbalanced data. AdaBoost and SMOTE optimize synthetic samples, implicitly modifying update weights and adjusting for skewed distributions. The typical AdaBoost technique uses too many system resources to prevent redundant or useless weak classifiers. To make process simple applied Adaptive PSO (APSO) to the SMOTE_AdaBoost results re-initialize of strategy to the optimize AdaBoost weak classifier coefficients. Four real imbalanced datasets on six classifiers—Naïve Bayes (NB), Random Forest (RF), Multi-layer Perception (MLP), Decision Tree (DT), and K-Nearest Neighbor (KNN)—verify the proposed multi-level strategy. The proposed strategy (APSO_SMOTE_AdaBoost) is applied to six classifiers’ and compared to SMOTE-PSO. The multi-level proposed strategy outperforms with standard approach in accuracy, precision, recall, sensitivity, and F-score.

The class imbalance is challenging issue in machine learning and data mining especially health care, telecom sector, agriculture sector, and many more (Zhu et al. in Pattern Recogn Lett 133:217–223, 2020; Thabtah et al. in Inf Sci 513:429–441, 2020). Imbalance of data samples across classes can arise as a result of human error, improper/unguided data sample selection, and so on (Tarekegn et al. in Pattern Recogn 118:107965, 2021). However, it is observed that applying imbalanced datasets to the data mining and machine learning approaches, it retains the biased in results which leads to the poor decision-making (Barella et al. in Inf Sci 553:83–109, 2021; Zhang et al. in ISA Trans 119:152–171, 2021; Ahmed and Green in Mach Learn Appl 9:100361, 2022). The primary motivation for this research is to explore and develop novel ensemble approaches for dealing with class imbalance and efficient way of retrieving synthetic data. In this paper, an ensemble method called IPSO-SMOTE-AdaBoost is developed to solve the class imbalance problem by combining the synthetic minority oversampling technique (SMOTE) (Gao et al. in Neurocomputing 74:3456–3466, 2011; Prusty et al. in Prog Nucl Energy 100:355–364, 2017), improved particle swarm optimization (PSO) (Yang et al. in J Electron Inf Technol 38:373–380, 2016), and AdaBoost. AdaBoost combined with SMOTE provides an optimal set of synthetic samples, thereby modifying the updating weights and adjusting for skewed distributions. The typical AdaBoost approach, on the other hand, consumes far too many system resources to avoid redundant or ineffective weak classifiers. With the proposed ensemble framework, IPSO-SMOTE-AdaBoost, parameters can be re-initialized to counter the concept of local optimum as well with the SMOTE that is boosted with AdaBoost method. The proposed method is validated using three datasets on six classifiers: extra tree (ET), naive Bayes (NB), random forest (RF), support vector machine (SVM), decision tree (DT), and K-nearest neighbor (KNN). After that, the IPSO-SMOTE-AdaBoost is compared to the existing SMOTE-PSO. The evaluation of proposed work is done with measures, namely accuracy, precision, recall, sensitivity, and F-score, and result shows that the proposed technique outperformed the usual PSO and SMOTE variations.

Density peak clustering (DPC) has shown promising results for many complex problems when compared with other existing clustering techniques. Inspite of many advantages, DPC suffers with lack of cluster centroids and cut-off distance identification. Cut-off distance is the prominent parameter used in the calculation of local density. The improper choice of cut-off distance leads to improper cluster results. Currently, the cut-off distance is selected using decision graph or delta density or knee point detection or silhouette score or kernel functions. The main problem with the above functions for selecting the cut-off distance in DPC is that they often rely on heuristic or visually subjective criteria, making the choice of the optimal cut-off distance challenging and potentially sensitive to data characteristics. By leveraging metaheuristic optimisation algorithms, the process of selecting the cut-off distance becomes less subjective and data-driven, potentially leading to improved clustering results in DPC. This motivated us to work on the choice of cut-off distance by the usage of remora optimisation algorithm (ROA). The cluster results are improved by the usage of remora in selection of reliable cut-off distance ((Formula presented.). The effectiveness of the updated DPC with ROA is evaluated by applying on the eight datasets and compared with K-means, traditional DPC, DPC merged with other optimisation results. The three parameters used here to check the quality of the cluster are homogeneity, completeness and silhouette analysis. ROA is new and built on the inspiration of remora which moves from one place to another using the sea fishes like shark, whale, sword fish, etc. It is clear from the results that DPC with ROA has produced the better homogeneity value of 0.807, completeness of 0.699 and silhouette analysis of 0.79 than the other clustering algorithms.

Clustering is a technique used to group similar datapoints/samples. Similar group of datapoints can be formed by using distance measure or by density. Density peak clustering (DPC) groups datapoints based on the density. This paper shows variations and improvements of DPC and also the performance of DPC over other clustering algorithms. This paper also addresses the problem in DPC with random selection of cut-off distance parameter(dc). Local density of the datapoint is calculated based on dc. The improper selection of dc leads to wrong clustering results. The issue in the random choice of dc is addressed by using gini index or Gaussian function to make a valid guess on dc.. Here we have chosen homogeneity, completeness, silhouette coefficient as the three parameters to compare results of DPC, DPC with gini index, DPC with gaussian function.

Recommender Systems has grown significantly over the last two decades. Memory-based Collaborative Filtering is part of RS and is a powerful technology that has been applied in several well-established commercial applications.However, memory-based collaborative filtering fails to capture the dynamic user opinions in a detailed perceptive since it uses a two-dimensional rating approach.However, multicriteria RS dominates memory-based collaborative filtering with the inclusion of multiple contexts. In addition, significant research has been done to predict user gratification. However, recent multicriteria recommender systemsfail to avoid the significant issues of the curse of dimensionalitydue to the lower number of ratings among multiple dimensions, leading to poor predictions. This paper proposes a new prediction recommender model on multicriteria recommender systems to predict user gratification with the memory-based user and item collaborative filtering approaches used to impute the missing contextsin multicriteria RS. In addition, various regression models were applied to overall and predicted overall ratings. The results indicate that item-item collaborative filtering with Ordinary Least Squares(OLS) regression in multicriteria RS exhibits low Root Mean Squared Error(RMSE), indicating the accurate predictions of user gratification.

Most growing real-time applications operate with high-dimensional data and require appropriate feature selection and prediction analysis technique. The regularization approach is one of the most widely used strategies for genomic data processing. The selection of features in high-dimensional data with substantially linked variables is a critical challenge. As a result, in this work, a unique approach is proposed for feature selection, wherein TLFM (Two-Level Filtering for Microarray Data), is identified for achieving an optimal gene data. Using the Information Gain (IG), each gene was prioritized from the early stages based on its value for classification. At the first level, a subset of candidate genes is created. Later, the redundant genes are filtered and the retained information genes from the subset (i.e., candidate genes) obtained from the previous step is done by using the Extended Lasso (EL) method. The real-time datasets have tested the proposed method against the standard methods. The results of the proposed study proven that the proposed method has produced better classification results with fewer genes.

Tweets based micro blogging is the most widely used social media to share the opinions in terms of short messages. Tweets facilitate business men to release the products based on the user interest which thereby produces more profits to their business. It also helps the government to monitor the public opinion which leads to better policies and standards. The large number of tweets on different topics are shared daily so, there is a need to identify trending topics. This paper proposes a method for automatic detection of hot topics discussed predominantly in social media by aggregating tweets of similar topics into manageable clusters. This produces hot topic detection irrespective of the current user location. A Modified Density Peak Clustering (MDPC) algorithm based hot topic detection is proposed. Local density of traditional Density Peak Clustering (DPC) is redefined by using the gaussian function in the calculation of dc (threshold distance). The traditional DPC considering some random value as dc (threshold distance) this gives a negative impact on the cluster formation thereby return inappropriate clusters. This can be solved by using the MDPC. The MDPC algorithm works by taking the cosine similarity between the tweets as the input and produces clusters of similar tweets. The cluster having a greater number of tweets is considered as hot topic which is frequently discussed by most of the users on twitter. Events 2012 dataset is collected with streaming API. This contains tweets from 2012 to 2016. The dataset consists of 149 target events and 30 million tweets. Experimental result shows that the proposed algorithm performs better than the traditional algorithms such as density peak clustering, K-means clustering, and Spectral clustering. It has produced the accuracy of 97%.

In the modern age and many prestigious applications use the recommendation method to play an important role. The system of recommendations collected apps, built a global village and provided enough information for development. This paper presents an overview of the approaches and techniques produced in the recommendation framework for collaborative filtering. Collaborative filtering, material and hybrid methods were the method of recommendation. In producing personalised recommendation the technique of collaborative filtering is particularly effective. There have been several algorithms over ten years of study, but no distinctions have been made between the various strategies. Indeed, there is not yet a widely agreed way to test a collaborative filtering algorithm. In this work we compare various literature techniques and review each one's characteristics to emphasise their key strengths and weaknesses.

In many applications, the feature selection plays an important role as only when we can get the best feature, we can bring out accurate results. The features selected must represent the entire dataset. Here we have chosen the whale optimization algorithm for feature extraction. To the whale optimization algorithm, we have added the convergence function and the fitness function. The fitness function is used to check the accuracy of the algorithm. Here we also used the Minkowski distance between feature and the cluster centroid to group the common features together. Grouping of common features is useful in many applications like applying a common methodology to the similar feature, spam detection, email classification. This paper has proposed an algorithm which extracts the features by checking its accuracy with fitness function and clusters the common features using Minkowski distance and k clusters.

Indian Premier League is a T20 League which was started in 2008 and now became the most irresistible T20 cricket carnival. Since the IPL has large popularity, predicting the results of it is really important and to be more effective. The Solution of predicting the results can be done with the help of Time Series Analysis and the Machine Learning Algorithms and Techniques which reduce the Domain Knowledge. Data Analysis has to be done by taking the historical data and need to draw some conclusions by applying Machine Learning Techniques. The solution of predicting the match must be effective since, there is a lot enthusiasm for IPL seasons and winners of that Season. Data Analytics are also used in Commercial Industries to draw the best conclusions. In this particular paper the parameters like Venue of the match, Win or Loss of the Toss, ball to ball details, Batsman Strike Rate were taken in to consideration for which the machine learning techniques were applied and the results are predicted. The Data Sets of past 7 years are taken with the above parameters and preprocessing is done for the data. The Machine Learning Algorithms that we used in here are Random Forest and Logistic Regression for predicting the accurate results. Before predicting, we need explore the data and analyze it to the extent.

The main aim of this paper is to detect anomaly in the dataset using the technique Outlier Removal Clustering (ORC) on IRIS dataset. This ORC technique simultaneously performs both K-means clustering and outlier detection. We have also shown the working of ORC technique. The datapoints which is far away from the cluster centroid are considered as outliers. The outliers affect the overall performance and result so the focus is on to detect the outliers in the dataset. Here, we have adopted the preprocessing technique to handle the missing data and categorical variable to get the accurate output. To select the initial centroid we have used Silhouette Coefficient.

In many applications, the feature selection plays an important role, as best feature can bring out the accurate results. The features selected must represent the entire dataset. Here we have considered 'Sequential Forward Selection' for feature extraction and used refined dragonfly algorithm to approach and to migrate from the best and worst features respectively. We improvised the conventional dragonfly algorithm by adding the convergence and fitness functions. To access the accuracy of the algorithm we introduced the fitness function. This paper has discussed about the general hunting behaviour of the dragonfly and dragonfly algorithm (DA) with convergence and fitness functions. A comparative study was shown for the best search agent position between modified DA and traditional DA, at the same time test function values of refined dragonfly algorithm (RDA) is compared with whale optimization algorithm (WOA) and Tornadogenesis Optimization algorithm (TOA). We have evaluated refined DA on the 23 benchmark function corresponding values are shown in experiment.

In this paper we have shown the behaviour of the new k-means algorithm. In k-means clustering first we take the ‘n’ number of item sets, then we group those item sets into the k clusters by placing the item set in the cluster with nearest mean. The traditional k-means clustering is completely depend on initial clusters and can be used only on spherical-shape clusters. The traditional k-means clustering uses the euclidean distance but in our paper we have replaced it with minkowski distance and combined with the Generalized Sequential Pattern algorithm (GSP algorithm) to find the frequent item sets in the sequential data stream. The GSP algorithm based on the frequent item sets, it traces the databases iteratively. The modified k-means clustering have reduce the complexity and calculations and the GSP algorithm has given the better result than any other algorithm to find the frequent item sets. The results show that this approach has given the better performance when compared to the traditional k means clustering.

A common threat for maximum computers today are due to malwares. In recent years attackers created different types of malwares which has become a challenge for many anti malware software. The malware companies have generated various forms of same malware. The different forms of the same malware will have similar functionality and same behaviour but with various representation. Here we cluster (group) the similar behaviour malware using number of distance measure. For clustering (group) malware samples we have many approaches which are published. There by different similarity measures are used but without thoroughly discussing their choice. Here we discuss about various similarity measure and their properties to get the accurate output. Our main focus is on behavioural features of malware and compare. Here we have used K means for clustering the malware samples.

In the field of medical applications, typically obtained medical images like X-ray, CT, MRI etc. consists of noise that reduces the visual quality of an image. Therefore, de-noising is essential during the image acquisition process. Though several methods are available for de-noising the image, the performance metrics of wavelets and threshold values to be used are not optimized for assessing the quality of an image. In this paper, DWT techniques with suitable threshold value and five objective quality metrics are used for de-noising the abnormal MRI brain speckle noise image. Quality metrics like Squared Error Mean (SEM), Peak Signal to Noise Ratio (PSNR), Structural content (SC), Structural Similarity Index Method (SSIM), and Absolute Mean Error (AME) are estimated for de-noised MRI brain image are discussed. The quality of the image is assessed depending on the metrics and wavelet threshold techniques.