H.3. Artificial Intelligence
Damianus Kofi Owusu; Christiana Cynthia Nyarko; Joseph Acquah; Joel Yarney
Abstract
Head and neck cancer (HNC) recurrence is ever increasing among Ghanaian men and women. Because not all machine learning classifiers are equally created, even if multiple of them suite very well for a given task, it may be very difficult to find one which performs optimally given different distributions. ...
Read More
Head and neck cancer (HNC) recurrence is ever increasing among Ghanaian men and women. Because not all machine learning classifiers are equally created, even if multiple of them suite very well for a given task, it may be very difficult to find one which performs optimally given different distributions. The stacking learns how to best combine weak classifier models to form a strong model. As a prognostic model for classifying HNSCC recurrence patterns, this study tried to identify the best stacked ensemble classifier model when the same ML classifiers for feature selection and stacked ensemble learning are used. Four stacked ensemble models; in which first one used two base classifiers: gradient boosting machine (GBM) and distributed random forest (DRF); second one used three base classifiers: GBM, DRF, and deep neural network (DNN); third one used four base classifiers: GBM, DRF, DNN, and generalized linear model (GLM); and fourth one used five base classifiers: GBM, DRF, DNN, GLM, and Naïve bayes (NB) were developed, using GBM meta-classifier in each case. The results showed that implementing stacked ensemble technique consisting of five base classifiers on gradient boosted features achieved better performance than achieved on other feature subsets, and implementing this stacked ensemble technique on gradient boosted features achieved better performance compared to other stacked ensemble techniques implemented on gradient boosted features and other feature subsets used. Learning stacked ensemble technique having five base classifiers on GBM features is clinically appropriate as a prognostic model for classifying and predicting HNSCC patients’ recurrence data.
H.3. Artificial Intelligence
Ali Rebwar Shabrandi; Ali Rajabzadeh Ghatari; Nader Tavakoli; Mohammad Dehghan Nayeri; Sahar Mirzaei
Abstract
To mitigate COVID-19’s overwhelming burden, a rapid and efficient early screening scheme for COVID-19 in the first-line is required. Much research has utilized laboratory tests, CT scans, and X-ray data, which are obstacles to agile and real-time screening. In this study, we propose a user-friendly ...
Read More
To mitigate COVID-19’s overwhelming burden, a rapid and efficient early screening scheme for COVID-19 in the first-line is required. Much research has utilized laboratory tests, CT scans, and X-ray data, which are obstacles to agile and real-time screening. In this study, we propose a user-friendly and low-cost COVID-19 detection model based on self-reportable data at home. The most exhausted input features were identified and included in the demographic, symptoms, semi-clinical, and past/present disease data categories. We employed Grid search to identify the optimal combination of hyperparameter settings that yields the most accurate prediction. Next, we apply the proposed model with tuned hyperparameters to 11 classic state-of-the-art classifiers. The results show that the XGBoost classifier provides the highest accuracy of 73.3%, but statistical analysis shows that there is no significant difference between the accuracy performance of XGBoost and AdaBoost, although it proved the superiority of these two methods over other methods. Furthermore, the most important features obtained using SHapely Adaptive explanations were analyzed. “Contact with infected people,” “cough,” “muscle pain,” “fever,” “age,” “Cardiovascular commodities,” “PO2,” and “respiratory distress” are the most important variables. Among these variables, the first three have a relatively large positive impact on the target variable. Whereas, “age,” “PO2”, and “respiratory distress” are highly negatively correlated with the target variable. Finally, we built a clinically operable, visible, and easy-to-interpret decision tree model to predict COVID-19 infection.
M. Rahimi; A. A. Taheri; H. Mashayekhi
Abstract
Finding an effective way to combine the base learners is an essential part of constructing a heterogeneous ensemble of classifiers. In this paper, we propose a framework for heterogeneous ensembles, which investigates using an artificial neural network to learn a nonlinear combination of the base classifiers. ...
Read More
Finding an effective way to combine the base learners is an essential part of constructing a heterogeneous ensemble of classifiers. In this paper, we propose a framework for heterogeneous ensembles, which investigates using an artificial neural network to learn a nonlinear combination of the base classifiers. In the proposed framework, a set of heterogeneous classifiers are stacked to produce the first-level outputs. Then these outputs are augmented using several combination functions to construct the inputs of the second-level classifier. We conduct a set of extensive experiments on 121 datasets and compare the proposed method with other established and state-of-the-art heterogeneous methods. The results demonstrate that the proposed scheme outperforms many heterogeneous ensembles, and is superior compared to singly tuned classifiers. The proposed method is also compared to several homogeneous ensembles and performs notably better. Our findings suggest that the improvements are even more significant on larger datasets.
M. Rezaei; H. Nezamabadi-pour
Abstract
The present study aims to overcome some defects of the K-nearest neighbor (K-NN) rule. Two important data preprocessing methods to elevate the K-NN rule are prototype selection (PS) and prototype generation (PG) techniques. Often the advantage of these techniques is investigated separately. In this paper, ...
Read More
The present study aims to overcome some defects of the K-nearest neighbor (K-NN) rule. Two important data preprocessing methods to elevate the K-NN rule are prototype selection (PS) and prototype generation (PG) techniques. Often the advantage of these techniques is investigated separately. In this paper, using the gravitational search algorithm (GSA), two hybrid schemes have been proposed in which PG and PS problems have been considered together. To evaluate the classification performance of these hybrid models, we have performed a comparative experimental study including a comparison between our proposals and some approaches previously studied in the literature using several benchmark datasets. The experimental results demonstrate that our hybrid approaches outperform most of the competitive methods.
P. Farzi; R. Akbari
Abstract
Abstract: Web service is a technology for defining self-describing objects, structural-based, and loosely coupled applications. They are accessible all over the web and provide a flexible platform. Although service registries such as Universal Description, Discovery, and Integration (UDDI) provide facilities ...
Read More
Abstract: Web service is a technology for defining self-describing objects, structural-based, and loosely coupled applications. They are accessible all over the web and provide a flexible platform. Although service registries such as Universal Description, Discovery, and Integration (UDDI) provide facilities for users to search requirements, retrieving the exact results that satisfy users’ need is still a difficult task since providers and requesters have various views about descriptions with different explanations. Consequently, one of the most challenging obstacles in the discovery task would be how to understand both sides, which is called knowledge-based understanding. This is of immense value for search engines, information retrieval tasks, and even NPL-based various tasks. The goal is to help recognize matching degrees precisely and retrieve the most relevant services more straightforward. In this research, we introduce a conceptual similarity method as a new way that facilitates discovery procedure with less dependency on the provider and user descriptions to reduce the manual intervention of both sides and being more explicit for the machines. We provide a comprehensive knowledge-based approach by applying the Latent Semantic Analysis (LSA) model to the ontology scheme - WordNet and domain-specific in-sense context-based similarity algorithm. The evaluation of our similarity method, done on OWL-S test collection, shows that a sense-context similarity algorithm can boost the disambiguation procedure of descriptions, which leads to conceptual clarity. The proposed method improves the performance of service discovery in comparison with the novel keyword-based and semantic-based methods.
Z. Shojaee; Seyed A. Shahzadeh Fazeli; E. Abbasi; F. Adibnia
Abstract
Today, feature selection, as a technique to improve the performance of the classification methods, has been widely considered by computer scientists. As the dimensions of a matrix has a huge impact on the performance of processing on it, reducing the number of features by choosing the best subset of ...
Read More
Today, feature selection, as a technique to improve the performance of the classification methods, has been widely considered by computer scientists. As the dimensions of a matrix has a huge impact on the performance of processing on it, reducing the number of features by choosing the best subset of all features, will affect the performance of the algorithms. Finding the best subset by comparing all possible subsets, even when n is small, is an intractable process, hence many researches approach to heuristic methods to find a near-optimal solutions. In this paper, we introduce a novel feature selection technique which selects the most informative features and omits the redundant or irrelevant ones. Our method is embedded in PSO (Particle Swarm Optimization). To omit the redundant or irrelevant features, it is necessary to figure out the relationship between different features. There are many correlation functions that can reveal this relationship. In our proposed method, to find this relationship, we use mutual information technique. We evaluate the performance of our method on three classification benchmarks: Glass, Vowel, and Wine. Comparing the results with four state-of-the-art methods, demonstrates its superiority over them.
G.3.9. Database Applications
M. Shamsollahi; A. Badiee; M. Ghazanfari
Abstract
Heart disease is one of the major causes of morbidity in the world. Currently, large proportions of healthcare data are not processed properly, thus, failing to be effectively used for decision making purposes. The risk of heart disease may be predicted via investigation of heart disease risk factors ...
Read More
Heart disease is one of the major causes of morbidity in the world. Currently, large proportions of healthcare data are not processed properly, thus, failing to be effectively used for decision making purposes. The risk of heart disease may be predicted via investigation of heart disease risk factors coupled with data mining knowledge. This paper presents a model developed using combined descriptive and predictive techniques of data mining that aims to aid specialists in the healthcare system to effectively predict patients with Coronary Artery Disease (CAD). To achieve this objective, some clustering and classification techniques are used. First, the number of clusters are determined using clustering indexes. Next, some types of decision tree methods and Artificial Neural Network (ANN) are applied to each cluster in order to predict CAD patients. Finally, results obtained show that the C&RT decision tree method performs best on all data used in this study with 0.074 error. All data used in this study are real and are collected from a heart clinic database.
G.3.5. Systems
M. Rezvani
Abstract
Cloud computing has become an attractive target for attackers as the mainstream technologies in the cloud, such as the virtualization and multitenancy, permit multiple users to utilize the same physical resource, thereby posing the so-called problem of internal facing security. Moreover, the traditional ...
Read More
Cloud computing has become an attractive target for attackers as the mainstream technologies in the cloud, such as the virtualization and multitenancy, permit multiple users to utilize the same physical resource, thereby posing the so-called problem of internal facing security. Moreover, the traditional network-based intrusion detection systems (IDSs) are ineffective to be deployed in the cloud environments. This is because that such IDSs employ only the network information in their detection engine and this, therefore, makes them ineffective for the cloud-specific vulnerabilities. In this paper, we propose a novel assessment methodology for anomaly-based IDSs for cloud computing which takes into account both network and system-level information for generating the evaluation dataset. In addition, our approach deploys the IDS sensors in each virtual machine in order to develop a cooperative anomaly detection engine. The proposed assessment methodology is then deployed in a testbed cloud environment to generate an IDS dataset which includes both network and system-level features. Finally, we evaluate the performance of several machine learning algorithms over the generated dataset. Our experimental results demonstrate that the proposed IDS assessment approach is effective for attack detection in the cloud as most of the algorithms are able to identify the attacks with a high level of accuracy.
H.3.2.3. Decision support
F. Moslehi; A.R. Haeri; A.R. Moini
Abstract
In today's world, most financial transactions are carried out using done through electronic instruments and in the context of the Information Technology and Internet. Disregarding the application of new technologies at this field and sufficing to traditional ways, will result in financial loss and customer ...
Read More
In today's world, most financial transactions are carried out using done through electronic instruments and in the context of the Information Technology and Internet. Disregarding the application of new technologies at this field and sufficing to traditional ways, will result in financial loss and customer dissatisfaction. The aim of the present study is surveying and analyzing the use of electronic payment instruments in banks across the country using statistics and information retrieved from the Central Bank and data mining techniques. For this purpose, firstly, according to the volume of transactions carried out and with the help of using the K-Means algorithm, a label was dedicated to any record; then hidden patterns of the E-payment instruments transaction were detected using the CART algorithm. The obtained results of this study enable banks administrators to balance their future policies in the field of E-payment and in the bank and customers’ interest's direction based on detected patterns and provide higher quality services to their customers.
H.6.4. Clustering
M. Manteqipour; A.R. Ghaffari Hadigheh; R. Mahmoodvand; A. Safari
Abstract
Grouping datasets plays an important role in many scientific researches. Depending on data features and applications, different constrains are imposed on groups, while having groups with similar members is always a main criterion. In this paper, we propose an algorithm for grouping the objects with random ...
Read More
Grouping datasets plays an important role in many scientific researches. Depending on data features and applications, different constrains are imposed on groups, while having groups with similar members is always a main criterion. In this paper, we propose an algorithm for grouping the objects with random labels, nominal features having too many nominal attributes. In addition, the size constraint on groups is necessary. These conditions lead to a mixed integer optimization problem which is not convex nor linear. It is an NP-hard problem and exact solution methods are computationally costly. Our motivation to solve such a problem comes along with grouping insurance data which is essential for fair pricing. The proposed algorithm includes two phases. First, we rank random labels using fuzzy numbers. Afterwards, an adjusted K-means algorithm is used to produce homogenous groups satisfying a cluster size constraint. Fuzzy numbers are used to compare random labels, in both observed values and their chance of occurrence. Moreover, an index is defined to find the similarity of multi-valued attributes without perfect information with those accompanied with perfect information. Since all ranks are scaled into the interval [0,1], the result of ranking random labels does not need rescaling techniques. In the adjusted K-means algorithm, the optimum number of clusters is found using coefficient of variation instead of Euclidean distance. Experiments demonstrate that our proposed algorithm produces fairly homogenous and significantly different groups having requisite mass.
H.6.3.2. Feature evaluation and selection
M. Imani; H. Ghassemian
Abstract
Feature extraction is a very important preprocessing step for classification of hyperspectral images. The linear discriminant analysis (LDA) method fails to work in small sample size situations. Moreover, LDA has poor efficiency for non-Gaussian data. LDA is optimized by a global criterion. Thus, it ...
Read More
Feature extraction is a very important preprocessing step for classification of hyperspectral images. The linear discriminant analysis (LDA) method fails to work in small sample size situations. Moreover, LDA has poor efficiency for non-Gaussian data. LDA is optimized by a global criterion. Thus, it is not sufficiently flexible to cope with the multi-modal distributed data. We propose a new feature extraction method in this paper, which uses the boundary semi-labeled samples for solving small sample size problem. The proposed method, which called hybrid feature extraction based on boundary semi-labeled samples (HFE-BSL), uses a hybrid criterion that integrates both the local and global criteria for feature extraction. Thus, it is robust and flexible. The experimental results with three real hyperspectral images show the good efficiency of HFE-BSL compared to some popular and state-of-the-art feature extraction methods.
J.10.3. Financial
G. Ozdagoglu; A. Ozdagoglu; Y. Gumus; G. Kurt Gumus
Abstract
Predicting financially false statements to detect frauds in companies has an increasing trend in recent studies. The manipulations in financial statements can be discovered by auditors when related financial records and indicators are analyzed in depth together with the experience of auditors in order ...
Read More
Predicting financially false statements to detect frauds in companies has an increasing trend in recent studies. The manipulations in financial statements can be discovered by auditors when related financial records and indicators are analyzed in depth together with the experience of auditors in order to create knowledge to develop a decision support system to classify firms. Auditors may annotate the firms’ statements as “correct” or “incorrect” to add their experience, and then these annotations with related indicators can be used for the learning process to generate a model. Once the model is learned and tested for validation, it can be used for new firms to predict their class values. In this research, we attempted to reveal this benefit in the framework of Turkish firms. In this regard, the study aims at classifying financially correct and false statements of Turkish firms listed on Borsa İstanbul, using their particular financial ratios as indicators of a success or a manipulation. The dataset was selected from a particular period after the crisis (2009 to 2013). Commonly used three classification methods in data mining were employed for the classification: decision tree, logistic regression, and artificial neural network, respectively. According to the results, although all three methods are performed well, the latter had the best performance, and it outperforms other two classical methods. The common ground of the selected methods is that they pointed out the Z-score as the first distinctive indicator for classifying financial statements under consideration.
H.6.3.2. Feature evaluation and selection
Maryam Imani; Hassan Ghassemian
Abstract
When the number of training samples is limited, feature reduction plays an important role in classification of hyperspectral images. In this paper, we propose a supervised feature extraction method based on discriminant analysis (DA) which uses the first principal component (PC1) to weight the scatter ...
Read More
When the number of training samples is limited, feature reduction plays an important role in classification of hyperspectral images. In this paper, we propose a supervised feature extraction method based on discriminant analysis (DA) which uses the first principal component (PC1) to weight the scatter matrices. The proposed method, called DA-PC1, copes with the small sample size problem and has not the limitation of linear discriminant analysis (LDA) in the number of extracted features. In DA-PC1, the dominant structure of distribution is preserved by PC1 and the class separability is increased by DA. The experimental results show the good performance of DA-PC1 compared to some state-of-the-art feature extraction methods.
Amir Mosavi
Abstract
Often in modeling the engineering optimization design problems, the value of objective function(s) is not clearly defined in terms of design variables. Instead it is obtained by some numerical analysis such as FE structural analysis, fluid mechanic analysis, and thermodynamic analysis, etc. Yet, the ...
Read More
Often in modeling the engineering optimization design problems, the value of objective function(s) is not clearly defined in terms of design variables. Instead it is obtained by some numerical analysis such as FE structural analysis, fluid mechanic analysis, and thermodynamic analysis, etc. Yet, the numerical analyses are considerably time consuming to obtain the final value of objective function(s). For the reason of reducing the number of analyses as few as possible our methodology works as a supporting tool to the meta-models. The research in meta-modeling for multiobjective optimization are relatively young and there is still much to do. Here is shown that visualizing the problem on the basis of the randomly sampled geometrical big-data of computer aided design (CAD) and computer aided engineering (CAE) simulation results, combined with utilizing classification tool of data mining could be effective as a supporting system to the available meta-modeling approaches. To evaluate the effectiveness of the proposed method a study case in 3D wing optimal design is given. Along with the study case, it is discussed that how effective the proposed methodology could be in further practical engineering design problems.
