Recommender Systems have become integral to most e-commerce applications and online platforms. The recommended suggestions heavily impact customer retention and business performance. One of the essential parameters in large-scale recommender systems is the time required to present a recommendation. The more time it takes, the more it loses the customer’s attention and interest. It is currently necessary for recommenders to be time-efficient and optimal. Collaborative filtering-based Matrix Factorization approaches have proven to be powerful for recommender systems. The standard approach uses the Singular Value Decomposition-based recommender systems with Gradient Descent optimizer and its advanced variants. These models provide good accuracy for recommenders. However, they are time-intensive. To alleviate these issues, the proximal gradient descent optimizer-based Nonnegative Matrix Factorization model is adapted for recommender systems to improve their performance in terms of time and accuracy. There has been no research on integrating proximal descent models in Nonnegative matrix factorization for recommender systems. These novel adaptations are analyzed with six other baseline recommender models on two datasets. The experimental analysis proves that these novel recommender models are the preferable choice for online recommenders and work well when the data is not smooth.

Recommender Systems have gained attraction since last two decades owing to its popularity of providing customers with the information of their choice, items they might purchase, movies they might watch and many preferences. This has led recommenders to penetrate almost all e-commerce and industrial applications. The Netflix Prize competition has shown the rising path to recommenders and Singular Value Decomposition (SVD) based factorization techniques. SVD and its variants are considered as one of the efficient approach for recommendation task, especially aligned for collaborative filtering task. However, these SVD approaches remains less interpretable and explainable, due to the presence of negative components in latent features of entities. In a realistic scenario, it is not possible to have negative components in the features. This leads to exploration of factorization techniques focused on positive components only. The non-negative counterpart of SVD, Non-negative Matrix Factorization (NMF) imposes the non-negativity constraint to features, allowing for more interpretable results. It has wide applicability ranging from Bioinformatics, topic modeling to recommenders. Till date, there have been several survey dedicated to core working of NMF. However, there was very limited literature available as to how different NMF works with recommender systems. There was no streamlined survey on NMF based recommender models that explores NMF in depth with advanced domains, case study and future directions. This article presents the insight on NMF based recommender models with focus on their mathematical modeling and taxonomy based on information they use, penalization term and optimization approach. The article offers the NMF based recommender model based on emerging approaches like Deep Learning, Federated Learning, Transformers, Explainability and Dynamicity. The article also brings up the open issues and challenges of NMF models. Further, the article presented an experimental analysis on NMF models with SVD and Neural Network model. The experimental analysis was covered on three benchmarked datasets of recommenders along with stability analysis. The results showcased the performance improvement brought by NMF in recommenders. Through this article, one can plunge into the dynamics of NMF for recommenders.

Graph Neural Networks (GNN) have been at cornerstone for social recommenders since a decade and have shown promising results. Existing state-of-the-art GNN models based on Graph Convolutional Networks (GCN), have seen a strong emergence in the recommenders, owing to its ability to make graph learning lighter and less complex. These models focus on learning the user and item embeddings through user-item graph and user-user social graph. This learning is accomplished by utilizing the implicit rating feedback and user-user trust information. However, they overlook the impact of the explicit rating feedback and bidirectional social trust. The explicit rating feedback provides the actual preference elicitation, while the bidirectional trust aids in strengthening the user ties, similar to the real-world user-to-user connection in social networking platforms. Based on this notion, we adapt the approach of using the explicit rating and bidirectional trust information into four GCN based social recommender models. The experimental analysis on these models with the benchmarked recommender datasets is carried out. The results show the dominance and performance improvement achieved by adding the explicit rating and bidirectional trust.

Reducing energy consumption and addressing environmental challenges such as climate change are driving the global shift toward renewable energy. This study investigates the application of Genetic Algorithm (GA) optimization to enhance the performance of neural network models for solar irradiance forecasting. We evaluate three architectures—Long Short-Term Memory (LSTM), Bidirectional LSTM (BiLSTM), and Gated Recurrent Unit (GRU)—both with and without GA-based hyperparameter tuning. GA is employed to systematically search the hyperparameter space, identifying configurations that maximise predictive accuracy. Experimental results demonstrate that GA optimization significantly improves model performance, with up to 60% reductions in Mean Squared Error (MSE) and Root Mean Squared Error (RMSE), and approximately 20% gains in R-squared (R2) scores. Among all configurations, GA-optimized LSTM models yield the most accurate predictions. This work highlights a novel integration of heuristic optimization with deep learning, offering a scalable and effective framework for renewable energy forecasting and other time-series prediction tasks.

This paper presents a systematic review and analysis of artificial intelligence (AI) based breast cancer diagnosis. The goal is to provide a helpful review of the literature for the breast cancer research community. The research papers are selected and shortlisted by conducting an extensive and exhaustive literature review through PRISMA. The artificial intelligence-based models, datasets, number of images, and problem type are reviewed with detailed performance analysis. The support vector machine (SVM) and convolutional neural network (CNN) are the most widely used AI models for breast cancer diagnosis. Digital database for screening mammography (DDSM) is the most used breast cancer research dataset. CNN and deep learning models require a more significant number of images for better performance in breast cancer research. It has been noticed from the review studies that the classification models such as SVM, extreme learning machine (ELM), CNN, and multi-layer perceptron (MLP) performed well as compared to the other existing machine and deep learning models.

Predicting seismic events with high accuracy remains a crucial challenge due to the complex nature of earthquake phenomena and the devastating impact they can have on human life and infrastructure. This study introduces a novel approach by integrating convolutional neural networks (CNN) with long short-term memory (LSTM) units to enhance the prediction of earthquake events. Utilizing a comprehensive dataset that includes 37,478 global earthquake instances, the model processes spatial and temporal features such as geographic location, time, magnitude, depth, and additional derived time series data. Through rigorous preprocessing, including label encoding of geographical locations and normalization of all relevant features, we prepare the dataset for effective machine learning modelling. The proposed CNN+LSTM model aims to capture both spatial dependencies through CNN layers and temporal sequences via LSTM layers, addressing the dynamic and multi-dimensional nature of seismic data. Preliminary results indicate improved accuracy in predicting the magnitude and likelihood of earthquake occurrences, offering promising implications for early warning systems and preparedness strategies. Future work will focus on refining the model through hyperparameter tuning and expanding the dataset to include more granular temporal features.

The rapid progress in machine learning techniques has significantly transformed healthcare which enables the simultaneous and accurate detection of multiple diseases. This paper delves into the application of diverse machine learning algorithms for multi-disease detection by using a comprehensive dataset which focuses on three diseases i.e. diabetes, gonorrhoea, and typhoid. The multi-disease dataset has been meticulously pre-processed and graphically visualized to discern patterns and represent diseases against emotional states/urges and critical feelings. Subsequently, a range of machine learning classifiers which includes logistic regression, Adaboost, random forest, support vector machine, CatBoost, Light Gradient Boosting Classifier, Naïve Bayes, XGBoost, KNN, and Decision Tree, are trained on this dataset. Their performance across these different classes is rigorously evaluated using various parameters such as accuracy, F1 score, recall, and precision. During execution, Adaboost emerged as the top performer, by achieving an impressive accuracy of 94.37% and maintaining a precision, recall, and F1 score of 0.94, which indicates its robustness in multi-disease detection.

The increased usage of social networks in day-to-day life has paved the path for social recommender models. The research in social recommender has seen a multi-directional approach in the last decade, among which the graph neural network-based approaches are most sought. Since the graph-based approaches can efficiently replicate the intrinsic behavior of the social networks. Social networks work on the notion of relationships between users, which is generally known as trust in recommender. Trust is the centric part of the social recommender models. Trust exploitation differs from model to model. However, trust in most models is used at a directed or propagated level only. In this article, we first provide the dynamics of trust and its terminologies, followed by state-of-the-art social recommenders that exploit these trust relations using graph-based models followed by their comparative analysis. At last, we provide the open issues and promising research direction of social recommender and how trust can be further infused in recommender. This article is the first such survey of social recommenders focusing on trust usage.

Growing N 2 O emissions, which are caused by the application of synthetic fertilizers in a variety of locations and crop types, are currently a key concern in the fight against greenhouse gas emissions. Utilizing activity and capacity factors, we aim to forecast N 2 O emission quantities from the soil. Our approach employs a hybrid algorithmic strategy, commencing with regression for primary prediction. Subsequently, ensembling techniques and gradient boosting regressor to capture complex data interactions. This sequential approach enhances prediction accuracy. This sequential approach facilitates the capture of complex interactions within the data, enhancing prediction accuracy. Through extensive analysis across varied agricultural landscapes, we seek to provide insights into N 2 O emission dynamics and contribute to sustainable agricultural practices. This research underscores the importance of integrating advanced analytics with traditional methods to address pressing environmental challenges.

In this quickly developing world, automatic currency identification and recognition are crucial tasks. Several financial institutions, such as banks and hardware-based devices such as vending machines and slot machines, play an essential role in all monetary unification fields. Accurate coin recognition is essential in various contexts, including vending machines, currency exchange, and archaeological research. However, the distinctive visual characteristics of Brazilian coins, including variations in size, color, and design, pose significant challenges for automated classification. Most of the existing currency recognition systems are based on the physical properties of the currencies, such as length, breadth, and mass. At the same time, image-based methods rely on other properties like color, shape, and edge. This paper presents a novel deep-learning framework tailored to classify Brazilian coins. Our proposed deep learning framework leverages state-of-the-art convolutional neural networks (CNNs) to address these challenges. We introduce a Repetitive Feature Extractor Convolution Neural Network (RFE-CNN) model to recognize the currency faster and accurately. Our framework employs a multi-stage approach for coin classification. First, a pre-processing module handles coin localization and image enhancement to mitigate variations in lighting and background. Next, an RFE-CNN-based feature extractor extracts discriminative features from the coin images. We explore transfer learning from pre-trained models to enhance the model’s generalization capability, given limited data availability. We used a comprehensive dataset of Brazilian coins, comprising various denominations, minting years, and conditions, to facilitate model training and evaluation. The dataset includes high-resolution images captured under diverse lighting and environmental conditions, ensuring robust model performance in real-world scenarios. In conclusion, our proposed deep learning framework offers a powerful and efficient solution for classifying Brazilian coins. The framework’s adaptability makes it a valuable tool for recognizing coins from other regions with similar visual diversity and variability challenges. The proposed model has achieved a classification accuracy of 98.34% for the classification of Brazilian coins.

With an emphasis on solar and wind energy, this study examines current advancements in renewable energy. It thoroughly investigates the difficulties in forecasting the production of solar and wind energy, considering weather dynamics, problems with data dependability, and constraints in current machine learning algorithms. The paper investigates several prediction models to address these issues, focusing on multilayer perceptron models and convolutional neural networks for forecasting solar irradiance. It presents a unique multivariate hybrid deep neural model that considers climate for estimating solar irradiance one hour in advance. In terms of wind energy, it emphasizes how crucial it is for prediction models to incorporate a variety of meteorological characteristics, supporting a comprehensive strategy that combines information from both sources using machine and deep learning. The study profoundly details a 30 MW hybrid power plant dataset from ibvogt Solar India Pvt. Ltd. A case study assesses decision trees, random forests, and extended short-term memory models, and a substantial training R2 value of 0.94 is obtained. The Bidirectional Long Short-Term Memory model's satisfactory testing R2 value of 0.45 and RMSE of 0.596 show skillful generalization and accuracy. The accuracy of statistical models related to hybrid energy varies. The necessity for sophisticated models for integrating renewable energy into intelligent networks is highlighted by identifying possibilities and problems in generating and forecasting renewable energy. The models' ability to accurately estimate solar and wind energy output and capture variation is validated by the R2 values.

Cervical cancer is a widespread type of cancer that primarily affects women. Detecting it early is crucial for successful treatment and recovery. However, detecting cervical cancer before it spreads or reaches an advanced stage poses a significant challenge for the medical industry. To address this issue, a study was conducted that utilized various machine-learning classification models to predict cervical cancer. The dataset was balanced using the Adaptive Synthetic Sampling approach, and principal component analysis was used to select the top 13 features. Performance was evaluated based on several metrics, including Accuracy Score, Precision Score, Recall Score, and F1 Score. The research aimed to enhance the existing model and provide a more efficient one. The study's results are discussed in detail in this article, and medical institutes can leverage these methods to predict cervical carcinoma.

Computational advancement like Artificial Intelligence (AI) and Internet of Things (IoT) has marked their importance in a variety of domains, healthcare being one such domain. The one major need of involving such advancements in healthcare is the data explosion. The modern healthcare uses these technologies in different medical area; in medical diagnostics to detect chronic disease like cancer and tumors, automated real-time monitoring of patient using sensors, medical image analysis and processing using deep learning (DL) and machine learning (ML), medical data analytics with the help of smart devices, treatment plan decision, development of medical equipment, drug discovery, robot-assisted surgeries, personalized healthcare, early disease detection, and many more. AI and IoT-enabled healthcare is currently at the niche stage and is evolving. The efficient usage of AI and IoT can change healthcare dynamics drastically in terms of decision making, predictive modelling, time consumption, and resource management. However, the inclusion of AI and IoT also tails its challenges like the accountability and trust of decisions made by AI, assurance of data quality which is transmitted through sensors, manpower for handling the AI- and IoT-enabled devices in healthcare environment, management of AI and IoT devices especially when linked with important medical equipment, consumption of energy and power by the devices that use sensor data, storage of data generated by the medical record, real-time management of the generated data, interpretation of data by machine, privacy, legal and security concerns of health records, human intervention, comfort of IoT devices for a patient and clinical staff and much more.With the current world scenario of pandemic and epidemics spreading around, it is of prime importance to dwell and see where these technological advancements stand in Healthcare Informatics and how they can be efficiently used for the futuristic healthcare sector. This chapter focuses on the proliferation of AI and IoT in various healthcare segments; clinical and non-clinical, along with the context of ML and DL and includes domains of healthcare benefit, AI and ML algorithms and techniques involved at various stages, use of IoT in healthcare, challenges faced, their limitations, and future scope.

Solar energy is going to play a crucial role in a green and clean environment by reducing pollution and global warming. Solar power generation forecasting is very important for the energy market and smart grids. The semi-structured data set was acquired from our university’s 1-Megawatt solar power facility. The data set was having twelve structured and one unstructured column. Manual entries concerning maintenance tasks and weather data are included in the unstructured column. With the help of weather parameters and maintenance operations, the unstructured column is utilized to generate new features. Random forest regressor, XG-boost, Naive Bayesian, univariate LSTM, and moving average models are applied to predict solar power generation using this data set. XG-boost and random forest regression models have less RMSE and MAE as compared to Naive Bayesian and univariate LSTM. The tree-based approaches work well with time-series data which has exogenous variables. The maintenance activities are also predicted in this research work. It is very difficult to predict the module cleaning in advance.

Sign Recognition System (SRS) is a technology that enables efficient communication between the deaf-mute community and the rest of society. Recent developments in machine learning and biomedical sensor technology have led to improvements in sign recognition systems. In this paper, we propose a sign recognition model that uses sensor data from a surface electromyography (sEMG) device and recognizes particular signs using a Stacked Bidirectional and Unidirectional LSTM (SBUL) architecture. The proposed model has significant social impact, as it facilitates communication among the deaf-mute community by recognizing signs using sEMG data. The neural network architecture proposed in this paper improves the performance of the model by considering temporal data dependencies between sEMG data and the layers of the proposed LSTM. This model provides information on temporal sequential data and offers an efficient method for SRS. Experimental results using sensor data demonstrate that the proposed SBUL network for sign recognition has an average recognition rate of approximately 96

This study aims to enhance biodiesel yield prediction's precision by applying the advanced Extreme Learning Machine (ELM) model. The optimisation algorithms include Cuckoo Search, Dragonfly Optimization, Firefly Optimization, and Fruitfly Optimization to improve the predictive efficiency of the ELM model. Two datasets are investigated: biodiesel from Ceiba pentandra oil and the bio-oil yield using a Box-Behnken method for the experiment design. Optimisation approaches find the optimal hidden size of the ELM model. The results obtained from the first dataset reveal that Dragonfly Optimization was found to be the most effective algorithm, exhibiting the lowest Root Mean Square Error (RMSE) of 0.098 for Dataset I and 0.000306 for Dataset II and a strong relationship between predicted and actual values with a high coefficient of determination (R-squared) of 0.99 in both datasets. Firefly Optimization achieved exceptional accuracy for the second dataset with a hidden size of 28. High coefficient of determination values, low RMSE, and Mean Square Error (MSE) values highlight the predictive precision of Firefly Optimization. The comparative analysis underscores the potential of optimisation methods to significantly enhance biodiesel yield estimates, with Firefly Optimization and Dragonfly Optimization emerging as standout performers for their respective datasets. This study contributes valuable insights into biodiesel production by showcasing the efficiency of optimised ELM models in predicting biodiesel and bio-oil outputs.

Solar energy forecasting has seen tremendous growth by using weather and photovoltaic (PV) parameters. This study presents new approach that predicts solar energy production by using the scheduled, unscheduled maintenance activities and weather data. The dataset is obtained from the 1MW solar power plant of PDEU (our university), which has 12 structured columns and 1 unstructured column with manual text entries about different scheduled and unscheduled maintenance activities, and weather conditions on the daily basis. The unstructured column is used to create new features by using Hash-Map, flag words and stop words. The solar power generation forecasting is formulated as a vector auto regression (VAR) optimization problem and total power generation forecasting is presented with the results of four different cases. The results have shown that the root mean square percentage error (RMSPE) in total power generation forecasting is less than 10% for different lag (p) values. The vector auto regression can forecast the unscheduled maintenance activities like Grid failure, Inverter Failure, scheduled maintenance activity like module cleaning, weather activity like cloudy along with total power generation forecasting for effective and efficient management of solar power plants. The power generation decay is different for all the PV sets which show the variations in the impacts of weather, aging and maintenance on the solar power plant. This research work has proven that the peaks of total power generation forecasting and prediction can be tracked in a better way by using daily unscheduled, scheduled maintenance activities and weather conditions.

In this paper, we have developed a hybrid model to predict fake news that includes clickbait detection as a parameter. The correlation between two different labels has been computed using a chi-square test. After establishing the correlation, the clickbait implementation was done on the heading of the dataset, and a fake news detection model has been executed on the content of the dataset. Then, the results of both the models were combined to generate a hybrid model through a regex equation. Our model is successful in enhancing the accuracy of the existing models by 1–2%. CCS Concepts • Artificial Intelligence • Natural Language Processing • Information Extraction.

Using econometric models or regression techniques, we can predict the price of a property based upon certain features. In this paper, we try to predict the sale price of the residential properties using advanced regression techniques and determine what features most affect the sale price. We have employed various feature engineering techniques to be applied to the raw data to make it productive for the machine learning models. An approach is discussed for understanding the black-box machine learning models, and interpretations are made using techniques like partial dependency plot. Such prediction models can be used in real-life scenarios to estimate the price of a property.

A novel technique to measure the resultant focal length of the thermal lens in an amplifier using collimated beam of the oscillator for copper vapour laser system demonstrated. The resultant focal length was estimated using ray tracing approach in a 40 W system (CVL-40). Effect of input electrical power and neon buffer gas pressure on the resultant focal length was studied. Role of equivalent thermal lens in master oscillator power amplifier (MOPA) configuration was evaluated. Compensation techniques for minimization of the thermal lensing effect for a MOPA chain were theoretically studied using ray tracing in an equivalent lensing model. Compensation of thermal lensing for an eight module MOPA chain with CVL-40 (38 mm bore, 1.8 m long, 40 W, 9 kHz) systems studied. An eight-module MOPA chain configuration with minimized thermal lensing based on optimized compensation techniques is proposed. Our study finds benefits for high power laser systems used in various industrial, research and development applications like micromachining, high speed photography, bio-medical applications and pumping dye lasers for spectroscopy.