Today, millions of people are suffering from diabetes which contributes to many other lethal diseases i.e., heart, kidney, and nerve damage. Diabetes mellitus is a chronic disease characterised by the body's inability to metabolise glucose, which could be life-threatening. Thus, several researchers have attempted to construct an accurate diabetes predictive model over the years. Big data analytics has played a vital role in healthcare by building predictive models for diabetes mellitus using various machine learning techniques. A large amount of data is collected which opened the opportunity to develop more complex, accurate predictions of the model. This paper aims to discuss various machine learning models to predict diabetes mellitus more accurately over the years. We have conducted a thorough review of the literature on predicting diabetes using PIMA and other datasets, which demonstrates how various machine-learning algorithms can be used to predict diabetes.
In the absence of an Amazigh knowledge database structuring biographical information, we propose a methodology to elaborate such database. In this paper, we give a brief overview of the concerned topics while introducing the respective approaches, and we present the process of making our biographical collection. Furthermore, we discuss the challenges we meet in finding the right balance between what we need to collect and what is available online. Our methodology focuses on biographical information extraction from press dispatches that have been annotated with named entities. It is based on linguistic patterns and lexical markers, while adopting 'local grammars' concept.
Social networks have become a customary news media source in recent times. However, the openness and unrestricted way of sharing information on social networks fosters spreading rumours which may cause severe damages economically, socially, etc. Motivated by this, our paper focuses on the rumour detection problem in Algerian Arabizi. Studying linguistic rules of Algerian Arabizi, we propose a lemmatiser and a parser for analysing and standardising the text to produce better rumour detection models. An approach for classifying rumours and news in social networks based on emotions' expression and users' positions is proposed. The experiments were done on many ngram representations where the best one has reached more than 94% of F-score. In addition to that this research deals with resources creation for Algerian Arabizi which is an under-resourced dialect. A corpus and several lexicons have been built, which can be the subject of other works dealing with this dialect.
The main objective of this study is to generate decision tree (DT) models/graphs (i.e., a type of supervised machine learning (ML) research method) through RapidMiner Studio (i.e., a popular visual platform for predictive analytics). The dataset used in the study contains attributes regarding the car accidents such as 'gender', 'casualty class', 'age group' and 'type of vehicle'. These are important features to decide whether the 'survival chance' of traffic accident patients would be 'high' or 'low'. Therefore, our goal is to apply 'DTs' for predicting the 'survival attribute' with the purpose of identifying high risk groups within the dataset. The resulting 'DTs' show that whenever the attribute 'gender' has the value 'male', and the attribute 'casualty class' has the value 'passenger', and the attribute 'gender' has the value 'male', and the attribute 'age group' has the value 'teenager'; then the 'survival chance' of the traffic/accident patient would be extremely 'low'.
Gait analysis compares the gait characteristics of people with health issues to those of a control group in order to detect gait abnormalities. This comparison is carried out by evaluating a number of gait parameters with discrete values. Gait data, on the other hand, is time-series data and must be assessed using a different approach. The purpose of this study was to develop a quantitative measure that takes into account time-series data for comparing the gait characteristics of two groups of individuals using clustering. The gait data were collected using an optical motion capture system. An adaptive density-peaks clustering technique with a shape-based similarity measure was employed to compare gait characteristics. The results demonstrate that the proposed adaptive density-peaks clustering technique, which employs dynamic derivative time wrapping distance measurement, outperforms three state-of-the-art clustering algorithms for comparing the gait characteristics using time-series gait data.
Designing an integrated supply chain network (SCN) by considering economic, social, and environmental aspects will increase its sustainability. This paper proposes a bi-objective mixed-integer linear programming model to design a sustainable closed-loop SCN considering location-inventory-routing problem, cross-docking scheduling, vehicles failure, inflation, routes reliability, impact of load on fuel consumption, discount on purchases from suppliers, supplier selection, and order allocation under uncertainty. The proposed bi-objective model minimises the total costs and maximises the created jobs, simultaneously. A scenario-based approach is used to cope with demand uncertainty while a fuzzy multi-objective solution approach is employed to convert the bi-objective model into a single-objective one. Finally, the performance of the proposed bi-objective mathematical model is evaluated using data from a wire and cable industry in Iran.
The main objective of the paper is to investigate the reasons why a lot of complaints (by customers) have been made against the quality of 'food delivery service' in one of the FitBox branches. During the COVID-19 pandemic, it is important for the FitBox customers to receive their (ordered) healthy food in-good-time to help them lose weight and get fit. To achieve the objective of the study, the whole dataset of the notorious branch was initially cleansed and then imported to the fluxicon disco platform, which is a process mining tool. Using several process mining techniques - such as automated process discovery (via frequency and time performance metrics), filtering (via follower, performance, endpoint and attribute metrics), clustering, process map animation/simulation and detailed statistics analytics - enabled us to find out the main reasons why the food delivery work has been piled up and handled improperly. The paper provides groundwork for future research.
Massive temporal data generated in different domains need to be analysed for decision-making by various applications. This time-series data holds significant contextual knowledge in the form of hidden events. The need for automatic identification of such events is apparent. The lack of effective pattern identification techniques for contextual events suggests the need for efficient event identification methods for various applications. The study aims to propose a contextual event identification methodology in temporal data using exploratory data learning. The exploratory learning algorithm identifies appropriate uncertainty limits in an iterative approach to get desired information gain. Audio music streams and standard text datasets are used to test the method and retrieve contextual events. The result shows a 1.8% improvement for text data compared with an LDA approach and 0.04% improvement in the mean reciprocal rate for music data. Contextual event identification is helpful for different decision-making tasks in machine learning. The proposed system is extendable in different domains such as network, financial or medical, where event identification of temporal data is essential.
This article proposes speakers' voice-based information (audio and video) retrieval systems, which combines speaker identification, speech recognition, and information retrieval algorithms. Information retrieval systems encompass system structure and a way to query the system for information retrieval. This article illustrates both, including how it is deployed on top of existing systems. The input to the system is a speaker voice sample and a text query. Based on the speaker's voice, the size of the corpus is reduced, and based on the text query, documents are retrieved and ranked. For the speaker identification, we used the LPC coefficient, for voice recognition, we used a Python speech recognition library, and for ranking, we used cosine similarity and TF-IDF. Other algorithms can replace any intermediate modules depending on the system, like crime investigation, news analysis, and lecture retrieval. We demonstrated the proposed method on simulated data generated from online websites.
In this paper, we address synonyms extraction through a simple algorithm based on equality operation of translated segments generated by four free of charge web machine translation systems namely: Google Translate , Promt , Babylon and Bing . The synonyms are classified into two kinds: strong synonyms with different vocabularies, and weak synonyms with the same vocabularies and different part of speech nature. The obtained results, using a dataset of ten theses abstracts, published into the web, seem to be very promising. Moreover, according to this study, it is advised that weak synonyms should receive more attention especially into synonyms substitution for machine translation evaluation.
This paper studies the decision-making process in purchasing used cars at a company. The company's main objective is to purchase cars that may be sold within 30 days. Currently, the decision is solely made based on the subjective judgment of a supervisor. Alternatively, utilising the data that has been collected by the company, a data mining approach is proposed to improve the decision-making process. Out of the 45 aspects of a car, 12 features are selected as being important using the contingency table method. Six data mining methods are applied. Support vector machine (SVM) prediction model performs the best. The SVM model provides an accuracy of 69.44% in predicting whether or not a used car would be successfully sold within the acceptable inventory days, i.e., 30 days. In contrast, the predictive accuracy of the current decision-making process is just around 50%.
Studying depression in Arabic social media has been neglected compared to other languages and the traditional way of dealing with depression (face-to-face medical diagnose) is not enough as the number of people that suffer from depression in Arabic communities increased dramatically. This paper proposes the first dataset to detect depressed users in Arabic social media. We pondered tweets from Twitter, pre-processed and converted it to a structured format. A notable advantage of the elaborated dataset is that it allows effective evaluation of machine learning algorithms for depression detection. We employ several classification algorithms such as deep neural network, logistic regression, multinomial Naïve Bayes, Bernoulli Naïve Bayes, AdaBoost, passive aggressive, nearest centroid, and linear SVC. The F-score, AUC, precision, and accuracy scores were selected as performance measures to compare algorithms, and the result showed that it is very challenging to classify Arabic tweets especially with the sparse nature of Twitter data.
Scheduling is an NP-hard problem which most universities are grappling with. For each academic semester, the timetabling process must be carried out regularly, which is an overwhelming and time-consuming activity. The main contribution of this study is developing an automated system based on a multi-agent (MA) approach and genetic algorithms (GA) to generate a university timetable. Three agents named capture agent (CA), processing agent (PA) and distributing agent (DA) have been worked collaboratively and cooperatively to develop the university timetable. The study has been applied in a real case study to perform the course schedules in the electrical and electronic engineering department of our university. The system implemented has considerably reduced the time and effort in the timetables realisation of our department from about ten days to only a few minutes. It has also significantly improved the quality of timetable by guaranteeing a satisfaction rate of over 95% of the constraints.
The fundamental job of sentiment analysis (SA) is to decide the sentiment polarity (positive or negative) of the text. It is a problematic task to take the sentiments in the document level sentences accurately. In the proposed system we developed sentiment analysis of book reviews using CNN with n-grams method by utilising two levels. In the first level, 'grouping of similar tagged words by semantic network' is completed taking pre-processed data utilising parts of speech (POS) Tagger from the datasets of books and reviewers by WuPalmer word similarity technique. In the second level, 'SA' is completed in two stages which are the training phase and testing phase by utilising deep learning approaches like convolutional neural networks (CNN) with n-gram method using document to vector (Doc2Vec) and distributed bag of words (DBOW) embedding. The proposed system 'CNN+Doc2Vec+DBOW+n-gram' divides the book reviews into positive or negative reviews with better accuracy results compared to existing methods.
The main goal of this research is the relationship between knowledge management and entrepreneurship with the role of social capital adjustment in Melli banks' employees. The sample was analysed using Cochran table and two-stage cluster sampling of the 117 employees of the Mashhad's Melli's bank. In this research, four standard questionnaires are used to measure the variables under study. The results of this study showed that there is a relationship between knowledge management and organisational entrepreneurship, and the social capital has a moderating effect on knowledge management and entrepreneurship.
Recently, the KSA has witnessed significant technical advances in the health sector, where local hospitals are using high-quality systems and technologies to serve patients. However, even with this great progress in healthcare systems, communication is still limited with other decision-makers in different sectors who need to access some health-related information to take the best decisions for serving patients. Therefore, this project aims to utilise the concept of electronic health governance (e-health governance) to build an automated system, which will help the health sector to know which common diseases are currently prevalent and facilitate the decision-making process by providing them with the necessary health information to help them provide the best service for patients in various fields. To do that this research will apply classification data mining techniques through using naïve Bayes classification algorithm; where this project aims to build a common diseases prediction system (CDPS) to work as syndromic surveillance system.
Electric vehicle (EV) sales are affected in many ways (especially in China), and there are few available sales forecasting models. The research was a decomposition and integration based on TEI@I methodology: the prediction model applied the principal component regression (PCR) analysis to deal with the linear relationship; then applied BP neural network and a support vector machine (SVM) to deal with the nonlinear relationship; and finally, they are all integrated together. Granger causality test and grey correlation degree are used to quantitatively analyse the factors affecting the sales of electric vehicles through mining consumer network data. The research results of EV models show that the Baidu search index lags behind for three months and is time-sensitive to the EV sales. Finally, taking the data of two car models as examples, it is found that the PCR-BP model and the PCR-SVM model have better prediction performance than the single model. It also provides an effective decision-making reference for similar product market prediction.
In data mining, feature selection plays an important role in finding the most important predictor variables (or features) that explain a major part of the variance of the response variable is a key to identify and build high performing models. In this proposed work, primary data is used to identify the normality/ abnormality of body weight. The missing data has been imputed by predictive mean matching (PMM) method. Efforts are made to reduce the dimensions of the data before classification using principal component analysis (PCA). The principal components obtained are passed as input to the supervised learning algorithm such as na
