Heart disease is still a major global health problem that needs early, accurate, and scalable prediction systems that can work in both clinical settings and remote healthcare delivery. This study suggests a better deep learning (DL) and IoT-based predictive method that can find heart diseases with high accuracy, explainability, and real-time use. A detailed comparison was made using ten different DL models, including ensemble DL, deep belief networks (DBN), tabular neural networks, XGBoost with neural network embedding, gradient-boosted neural networks (GBNN), multilayer perceptron (MLP) with feature engineering, long short-term memory (LSTM), residual neural network (ResNet), neural factorization machines (NFM), and recurrent neural networks (RNN). The classifiers were trained and tested on three sets of data: a public set with 685 samples, a private set with 355 samples, and a combined set with 1040 samples, using different cross-validation methods to confirm that the model works well. To improve prediction accuracy and speed up calculations, we used three statistical methods—ANOVA, Mutual Information (MI), and the chi-square test—to select the most important features. These methods helped isolate the most influential health indicators, reducing noise and dimensionality in the input space. Because medical datasets often have uneven class distributions, the SMOTE was used to create new samples from less common classes, helping to reduce the chance of biased learning. Among all models tested, the GBNN demonstrated superior performance across all datasets. Specifically, it achieved 96.35% accuracy, 94.83% sensitivity, 97.47% specificity, 96.49% precision, 95.65% F1 score, and 97.12% AUC on the public dataset; 97.18% accuracy, 100% sensitivity, 95.35% specificity, 93.33% precision, 96.55% F1 score, and 98.42% AUC on the private dataset; and 99.52% accuracy, 98.84% sensitivity, 100% specificity, 100% precision, 99.42% F1 score, and 99.33% AUC on the combined dataset. We used the SHAP method to make sure the system's predictions could be understood and trusted by clinicians. This explainable AI technique offers information about individual feature contributions for each prediction, helping clinicians and users understand the rationale behind classification outcomes and build confidence in the system’s decisions. As a final application layer, the top achievers integrated the DL into a mobile application that enables users to input how they feel and receive immediate forecasts of potential heart disease issues.
Suicide is a crucial reason for mortality and morbidity in Iran and worldwide. However, several organizations gather information on suicide and suicide attempts, yet there is substantial misperception regarding how to describe the phenomenon. In this study, first, a literature review was made to achieve a thorough overview of the related items of suicide. Then, the data items included from the literature review were analysed using a two-round Delphi study with content validation by an expert panel. Afterward, the suicide surveillance system was established based on the confirmed MDS, and ultimately, its recital was assessed by involving the end-users. The established: MDS was separated into administrative and clinical sections. Then, a web-based system with modular and layered architecture was developed based on the derived MDS. Finally, to evaluate the developed system, a survey of 150 persons of end-users was performed. The use of a consistent dataset with standardized values will improve scientific collaboration and other communications in this important area. The integration of information systems at the regional and national levels to data from suicide prevention programs makes it possible to assess the long-term outcomes and evolutions of interventions to prevent suicide.
Missing Data (MD) a substantial challenge that faces researchers when applying Data Mining (DM) techniques, notably when it comes to medical datasets. In the light of the different DM techniques, it is important to identify which ones are the more suitable to apply in the presence of MD in a given context. This paper evaluates and compares the impacts of different imputation techniques on the accuracy of DM techniques in the field of breast cancer classification. In fact, the paper investigates the use of three statistical imputation techniques: Mean, Median and Expectation-Maximization, three machine learning imputation techniques: K nearest neighbour, Regression Tree and Support Vector Regression, and forty-two heterogeneous ensembles based on these six single techniques. To evaluate these imputation techniques, five classifiers were assessed: Random Forest, Decision Tree (C4.5), Case-Based Reasoning, Support Vector Machines, and Multi-Layer Perceptron, in terms of three criteria: balanced accuracy rate, Kappa, and Area Under Curve, and over two Wisconsin breast cancer datasets. Moreover, we used: (1) the Scott-Knott (SK) clustering technique to cluster the 135 (27 imputation techniques * 5 classifiers) classifiers based on balanced accuracy values, and (2) the Borda count voting method based on the three criteria to rank the best SK classifiers. The findings showed that ML ensembles and hybrid ensembles achieved the highest performance in imputation compared to single techniques. Moreover, among single imputation techniques no technique guarantees the best classification results.
While the whole world struggles with the COVID-19 pandemic, there are many different measures taken by countries. In this sense, the distribution of free masks to citizens between the ages of 20-65 in Turkey is one of the important measures taken against to spread of the pandemic. This distribution process is carried out through pharmacies and people can obtain their masks from any pharmacy in their area of residence. However, this situation may cause some pharmacies to be very busy, and thus social distance cannot be maintained and the health and safety of the people may be threatened. In this paper, we aim to prioritize pharmacies so that only determined pharmacies in certain regions perform mask distribution process to prevent virus transmission. For this purpose, Esenler district is taken into consideration for a pilot study which is one of the risky regions in terms of virus spread in Istanbul, Turkey. Multi-criteria decision-making approach (MCDM) is used because of the necessity of handling many factors in decision-making process and the contradiction of evaluation factors in the prioritization of pharmacies. In order to best model the uncertainty in the decision process, the MCDM approach is applied in a fuzzy environment. In addition, spherical fuzzy AHP and VIKOR MCDM approaches are first used as a hybrid method in this paper which means the study also includes novelty in terms of the methodology adopted. As a result of spherical fuzzy multi-criteria analysis, the pharmacies that need to provide free mask distribution in the Esenler region have been successfully identified. Sensitivity analyses are also conducted to test the results' robustness by observing the change in results according to changing input parameters, and validity by employing different MCDM methods.
We suggested a new 3D graphical representation technique for protein sequences in this research. Our method used physicochemical property of twenty amino acid as well as BLOSUM62 matrix. This evolutionary information gives perceptive visualization. For analysis of protein similarity a particular vector from the 3D graphical representation curves is also formed. Our proposed method is verified on two datasets first on 8 different species ND6 protein sequences second on ND5 protein sequences gained from 22 different species. On the same dataset the phylogeny achieved by our method is agreed with previous researches. Our approach does not need aligning of protein sequences as done in old alignment methods.
With the architecture, Engineering and Construction (AEC) industry representing a significant share of global energy consumption and greenhouse gas (GHG) emissions, developing sustainable design and reducing buildings environmental impacts has become a priority over the past decades. When it comes to optimizing a buildings environmental impacts, there are many factors that affect the energy cost of the building either individually or in connection with others. As a result, it’s quite often that some of these attributes are overlooked due to their insignificance or in order to facilitate the analysis. With our environment constantly going through changes it is reasonable that the construction industry should also aim to adapt to these changes and make use of them. However, the role of local climate features and its effects on a buildings energy output is often so neglected. This research aims to consider the role of climatic attributes and local weather characteristics of a building by using BIM-LCA integration techniques, and see how it affects that buildings energy performance. Identical models featuring different climatic values were developed and the respective energy optimization results for different models were compared. An integrated BIM-LCA framework was also used to develop and assess the model’s energy performance before and after changing the wall systems in different regions and results were compared. Finally the CHG emissions for before and after the change were calculated. It is witnessed that by using the proper equipment and construction materials, that matches the respective climate, up to 28% of the buildings energy consumption during the operational phase, can be saved. In terms of energy savings, although, an insulated wall system is recommended, it was witnessed that adding the layer can contribute to increasing the GHG emissions of the model up to more than 3%. Moreover, this work develops a prototype to validate the results and also outlines a framework to automate the preparation of the Bill of Quantities (BoQ).
Aim: In this study, the mediators of the relationship between colleague solidarity and job performance for educators were examined. Rationale: One of the distinguishing features of educational institutions is that the learning and teaching processes, which are the core process of the organization, are mostly carried out by expert educators alone and independently in the classroom. This state of independence and loneliness causes colleague solidarity to emerge as a more important problem for educational institutions when compared to other institutions. Method: With the data collected from a total of 766 participants working in the education institutions, the mediator roles of the three individual variables, (work engagement, self-efficacy and thriving) and the three institutional variables (colleague relations, organizational climate and administrative support) between the colleague solidarity and the job performance of educators are examined. Findings: It has been determined that institutional and individual variables have a mediating effect in the relationship between educators' colleague solidarity and job performance. It was observed that institutional variables had a weaker mediating effect than individual variables in this effect. This result shows that the unique nature of the education profession reveals a different structure in terms of the variables examined in the research.
Innovations in technology emerged with digitalization affect all sectors, including supply chain and logistics. The term “digital supply chain” has arisen as a relatively new concept in the manufacturing and service sectors. Organizations planning to utilize the benefits of digitalization, especially in the supply chain area, have uncertainties on how to adapt digitalization, which criteria they will evaluate, what kind of strategies should be developed, and which should be given more importance. Multi-criteria decision making (MCDM) approaches can be addressed to determine the best strategy under various criteria in digital transformation. Because of the need to capture this uncertainty, fermatean fuzzy sets (FFSs) have been preferred in the study to widen the definition domain of uncertainty parameters. Interval-valued fermatean fuzzy sets (IVFFSs) are one of the most often used fuzzy set extensions to cope with uncertainty. Therefore, a new interval-valued fermatean fuzzy analytic hierarchy process (IVFF-AHP) method has been developed. After determining the main criteria and sub-criteria, the IVFF-AHP method has been used for calculating the criteria weights and ranking the alternatives. By determining the most important strategy and criteria, the study provides a comprehensive framework of digital transformation in the supply chain.