Pub Date : 2021-11-22DOI: 10.1109/ict4da53266.2021.9672226
Hussen Yesuf Ali, Sun Goulin, Abegaz Mohammed Seid
In a wireless network infrastructure, the initial synchronization process primarily decides whether to send or receive data between a device and base station. This process is usually powered by a random access (RA) mechanism to share and allocate radio resources dynamically. Over the past years, telecommunication industry has witnessed a massive growth in the Internet of Things (IoT) technologies which continue to be rolled out around the world with different services and having a variety of requirements. However, when massive IoT (mIoT) devices attempt to access the network over a limited number of Random Access Channel (RACH) resources within a time frame, the network becomes overloaded, leading to a low performance of human to human (H2H) communication and Quality of Services (QoS) may not be assured. To solve the above problems, we propose a dynamic resource slicing and access class barring (ACB) mechanism using deep reinforcement learning (DRL) for a new RACH scenario to control and manage the resource dynamically. Simulation results prove that our proposed technique provides a fair RACH resource allocation for each class according to the available radio resource.
{"title":"Autonomous RACH Resource Slicing for Heterogeneous IoT Devices Communication Using Deep Reinforcement Learning","authors":"Hussen Yesuf Ali, Sun Goulin, Abegaz Mohammed Seid","doi":"10.1109/ict4da53266.2021.9672226","DOIUrl":"https://doi.org/10.1109/ict4da53266.2021.9672226","url":null,"abstract":"In a wireless network infrastructure, the initial synchronization process primarily decides whether to send or receive data between a device and base station. This process is usually powered by a random access (RA) mechanism to share and allocate radio resources dynamically. Over the past years, telecommunication industry has witnessed a massive growth in the Internet of Things (IoT) technologies which continue to be rolled out around the world with different services and having a variety of requirements. However, when massive IoT (mIoT) devices attempt to access the network over a limited number of Random Access Channel (RACH) resources within a time frame, the network becomes overloaded, leading to a low performance of human to human (H2H) communication and Quality of Services (QoS) may not be assured. To solve the above problems, we propose a dynamic resource slicing and access class barring (ACB) mechanism using deep reinforcement learning (DRL) for a new RACH scenario to control and manage the resource dynamically. Simulation results prove that our proposed technique provides a fair RACH resource allocation for each class according to the available radio resource.","PeriodicalId":371663,"journal":{"name":"2021 International Conference on Information and Communication Technology for Development for Africa (ICT4DA)","volume":"115 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2021-11-22","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"124827279","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2021-11-22DOI: 10.1109/ict4da53266.2021.9672233
Tinsae Tadesse, Ketema Adere
The generic Mobile Ad-hoc Network protocols do not measure and use node quality indicators, such as remaining energy, in the route selection process. Recent Researches have included these measurements in the route acquisition and selection process. However; separately measuring these parameters can lead to larger packet size as the number of measured parameter increase. Our objectives are 1) to propose a metric called weight that generalizes the quality of nodes, and use it in the best route selection process of the Ad hoc On-Demand Distance Vector (AODV) protocol. 2) To compare the two path quality measurement methods, namely- the product and summation methods. The paper presents mathematical proofs of why the multiplication method is better than the summation and simulation-based comparison of the two methods. The simulated route-request and reply packets of the AODV protocol are modified to use the generalized weight metric in the routing process. Simulation results show that the summation method produces higher delay and packet delivery than both the product method and the AODV protocols.
{"title":"Comparison of Two End-to-End Path Quality Measurement Methods in Mobile Ad-hoc Networks","authors":"Tinsae Tadesse, Ketema Adere","doi":"10.1109/ict4da53266.2021.9672233","DOIUrl":"https://doi.org/10.1109/ict4da53266.2021.9672233","url":null,"abstract":"The generic Mobile Ad-hoc Network protocols do not measure and use node quality indicators, such as remaining energy, in the route selection process. Recent Researches have included these measurements in the route acquisition and selection process. However; separately measuring these parameters can lead to larger packet size as the number of measured parameter increase. Our objectives are 1) to propose a metric called weight that generalizes the quality of nodes, and use it in the best route selection process of the Ad hoc On-Demand Distance Vector (AODV) protocol. 2) To compare the two path quality measurement methods, namely- the product and summation methods. The paper presents mathematical proofs of why the multiplication method is better than the summation and simulation-based comparison of the two methods. The simulated route-request and reply packets of the AODV protocol are modified to use the generalized weight metric in the routing process. Simulation results show that the summation method produces higher delay and packet delivery than both the product method and the AODV protocols.","PeriodicalId":371663,"journal":{"name":"2021 International Conference on Information and Communication Technology for Development for Africa (ICT4DA)","volume":"21 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2021-11-22","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"123551414","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2021-11-22DOI: 10.1109/ict4da53266.2021.9672214
Setegn Asnakew Kasegn, S. Abebe
Identifier names are used to represent concepts in the source code. Concise and consistent identifier names are crucial to program comprehension. Identifier names reduce the effort to understand the software, support software maintenance and improve source code quality. Despite these benefits, many software systems are known to have meaningless and inconsistent identifier names. One of the reasons that lead to inconsistent identifier names is lack of knowledge of identifier names already used to represent concepts in the software. To address this problem, this study proposes a new approach to automatically suggest part of identifier name. The approach aims to use spatial locality to identify and suggest next terms given identifier name prefix. Spatial locality, in this context, refers to the use of terms in close proximity of documents related to the software system. The performance of our proposed approach is evaluated using six open source software systems. The evaluation result shows that the spatial locality based approach suggests part of identifier names correctly with an average precision of 83.2% and average mean reciprocal rank (MRR) of 25.5%. Of the top four correct suggestions, more than half are ranked in the first and second place.
{"title":"Spatial Locality Based Identifier Name Recommendation","authors":"Setegn Asnakew Kasegn, S. Abebe","doi":"10.1109/ict4da53266.2021.9672214","DOIUrl":"https://doi.org/10.1109/ict4da53266.2021.9672214","url":null,"abstract":"Identifier names are used to represent concepts in the source code. Concise and consistent identifier names are crucial to program comprehension. Identifier names reduce the effort to understand the software, support software maintenance and improve source code quality. Despite these benefits, many software systems are known to have meaningless and inconsistent identifier names. One of the reasons that lead to inconsistent identifier names is lack of knowledge of identifier names already used to represent concepts in the software. To address this problem, this study proposes a new approach to automatically suggest part of identifier name. The approach aims to use spatial locality to identify and suggest next terms given identifier name prefix. Spatial locality, in this context, refers to the use of terms in close proximity of documents related to the software system. The performance of our proposed approach is evaluated using six open source software systems. The evaluation result shows that the spatial locality based approach suggests part of identifier names correctly with an average precision of 83.2% and average mean reciprocal rank (MRR) of 25.5%. Of the top four correct suggestions, more than half are ranked in the first and second place.","PeriodicalId":371663,"journal":{"name":"2021 International Conference on Information and Communication Technology for Development for Africa (ICT4DA)","volume":"1 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2021-11-22","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"130801407","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2021-11-22DOI: 10.1109/ict4da53266.2021.9672241
Melaku Alelign, Tesfamariam M Abuhay, Adane Letta, Tizita Dereje
In 2018, more than 821 million undernourished people were registered all over the world. Of these, 239 million were in Sub-Saharan Africa. The numbers are particularly high in Ethiopia, Kenya, Somalia, and South Sudan. The determinant factors of food insecurity in Ethiopia are multidimensional encompassing climate change, civil conflicts, natural disasters, and social norms. This study, hence, aims to identify risk factors and predict food security status at household level in North West Ethiopia using supervised machine learning techniques. To this end, a dataset was gathered from the Dabat Health and Demographic Surveillance and statistically interesting risk factors were identified using logistics regression at a threshold level of p<0.05. Three experiments were also conducted using random forest, support vector machine and decision tree (C4.5) to predict food security status at household level and the performance of each model was evaluated using accuracy, precision, recall, and f1- measure. As a result, the C4.5 algorithm is selected as the best appropriate supervised machine learning algorithm with 97.23% of recall, 91.58% of accuracy, 80.97% of f1-measure, and 69.38% of precision. Family size, level of education, age of the household head, number and types of communication media, numbers of livestock, cultivated land size, access to credit, and access to irrigation are some of the risk factors of food security.
{"title":"Identifying Risk Factors and Predicting Food Security Status using Supervised Machine Learning Techniques","authors":"Melaku Alelign, Tesfamariam M Abuhay, Adane Letta, Tizita Dereje","doi":"10.1109/ict4da53266.2021.9672241","DOIUrl":"https://doi.org/10.1109/ict4da53266.2021.9672241","url":null,"abstract":"In 2018, more than 821 million undernourished people were registered all over the world. Of these, 239 million were in Sub-Saharan Africa. The numbers are particularly high in Ethiopia, Kenya, Somalia, and South Sudan. The determinant factors of food insecurity in Ethiopia are multidimensional encompassing climate change, civil conflicts, natural disasters, and social norms. This study, hence, aims to identify risk factors and predict food security status at household level in North West Ethiopia using supervised machine learning techniques. To this end, a dataset was gathered from the Dabat Health and Demographic Surveillance and statistically interesting risk factors were identified using logistics regression at a threshold level of p<0.05. Three experiments were also conducted using random forest, support vector machine and decision tree (C4.5) to predict food security status at household level and the performance of each model was evaluated using accuracy, precision, recall, and f1- measure. As a result, the C4.5 algorithm is selected as the best appropriate supervised machine learning algorithm with 97.23% of recall, 91.58% of accuracy, 80.97% of f1-measure, and 69.38% of precision. Family size, level of education, age of the household head, number and types of communication media, numbers of livestock, cultivated land size, access to credit, and access to irrigation are some of the risk factors of food security.","PeriodicalId":371663,"journal":{"name":"2021 International Conference on Information and Communication Technology for Development for Africa (ICT4DA)","volume":"9 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2021-11-22","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"125675066","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2021-11-22DOI: 10.1109/ict4da53266.2021.9672218
Mohammed Abebe, Özlem Aktaş, Süleyman Sevinç
The development of AI-based medical change tracking and impact analysis tools can have a beneficial effect on a patient's recovery in real-time. The study presents a system for patient medical change tracking and impact analysis using machine learning, particularly, principal component analysis and Bayesian structural networks. We found that the proposed system achieved an acceptable statistical significance level for all the patient data tested. Moreover, in cases where there are spurious changes due to extra missing values and/or newly administered medical tests causing the change, the causal impact analysis was able to capture them as bogus. Consequently, we can say that the proposed system can potentially offer real-time monitoring and tracking of patients for the clinicians. In addition, we believe that the approach provides a promising future in interpreting large quantities of patient data for establishing cause-effect relationships for critically ill patients.
{"title":"An Early Warning System for Evaluating Effects of Medical Treatment using Machine Learning","authors":"Mohammed Abebe, Özlem Aktaş, Süleyman Sevinç","doi":"10.1109/ict4da53266.2021.9672218","DOIUrl":"https://doi.org/10.1109/ict4da53266.2021.9672218","url":null,"abstract":"The development of AI-based medical change tracking and impact analysis tools can have a beneficial effect on a patient's recovery in real-time. The study presents a system for patient medical change tracking and impact analysis using machine learning, particularly, principal component analysis and Bayesian structural networks. We found that the proposed system achieved an acceptable statistical significance level for all the patient data tested. Moreover, in cases where there are spurious changes due to extra missing values and/or newly administered medical tests causing the change, the causal impact analysis was able to capture them as bogus. Consequently, we can say that the proposed system can potentially offer real-time monitoring and tracking of patients for the clinicians. In addition, we believe that the approach provides a promising future in interpreting large quantities of patient data for establishing cause-effect relationships for critically ill patients.","PeriodicalId":371663,"journal":{"name":"2021 International Conference on Information and Communication Technology for Development for Africa (ICT4DA)","volume":"46 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2021-11-22","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"122992010","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2021-11-22DOI: 10.1109/ict4da53266.2021.9672239
Shanko Chura Aredo, Y. Negash, Y. Wondie, Feyisa Debo, Rajaveerappa Devadas, Abreham Fekadu
Communication via millimeter-wave (mm-wave) has grown in favor as an alternative to the existing radio mobile communication technology for providing high gigabit-per-second data speeds. Because of the millimeter wave's short wavelengths, a large number of antennas may be placed in a compacted physical dimension to make a greater aperture and obtain a substantial gain in antenna arrays. Dirty paper coding (DPC) beamforming efficiently cancels the interference that the transmitter is aware of, resulting in increased capacity, energy, and spectral efficiency of mmWave enabled massive MIMO connectivity. However, the use of this techniques leads to complexity due to successive interference cancellation at detection when the number of users grow large. In this paper, a cooperative processing based multi-user precoding is presented for down-link mm-wave massive MIMO systems and thus cooperative precoding be implicitly designed by considering the digital beamforming solution which is obtained from the dirty paper after comparing the output of linear precoding schemes. The precoders are chosen and the system total rate is calculated based on the difficulty of detections based on user densities within virtual cells. Simulation results show that the proposed approach outperforms traditional digital beamforming in terms of sum rate.
{"title":"Density Aware Cooperative Precoding Technique for Massive MIMO Systems","authors":"Shanko Chura Aredo, Y. Negash, Y. Wondie, Feyisa Debo, Rajaveerappa Devadas, Abreham Fekadu","doi":"10.1109/ict4da53266.2021.9672239","DOIUrl":"https://doi.org/10.1109/ict4da53266.2021.9672239","url":null,"abstract":"Communication via millimeter-wave (mm-wave) has grown in favor as an alternative to the existing radio mobile communication technology for providing high gigabit-per-second data speeds. Because of the millimeter wave's short wavelengths, a large number of antennas may be placed in a compacted physical dimension to make a greater aperture and obtain a substantial gain in antenna arrays. Dirty paper coding (DPC) beamforming efficiently cancels the interference that the transmitter is aware of, resulting in increased capacity, energy, and spectral efficiency of mmWave enabled massive MIMO connectivity. However, the use of this techniques leads to complexity due to successive interference cancellation at detection when the number of users grow large. In this paper, a cooperative processing based multi-user precoding is presented for down-link mm-wave massive MIMO systems and thus cooperative precoding be implicitly designed by considering the digital beamforming solution which is obtained from the dirty paper after comparing the output of linear precoding schemes. The precoders are chosen and the system total rate is calculated based on the difficulty of detections based on user densities within virtual cells. Simulation results show that the proposed approach outperforms traditional digital beamforming in terms of sum rate.","PeriodicalId":371663,"journal":{"name":"2021 International Conference on Information and Communication Technology for Development for Africa (ICT4DA)","volume":"35 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2021-11-22","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"126167274","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Fault-based testing is a powerful technique to ensure the quality of software by evaluating the efficacy of the test suits and also used to check the thoroughness of testing performed by other software testing techniques. However, it is very complicated and computationally expensive testing method. Literature shows that there is a tremendous effort to give formal solutions and heuristics methods. Recently, state-of-the-art approaches based on hybrid optimization techniques have been proven to be suitable for cost effective results. This work implements and presents a multi-objective novel hybrid method by combining Backtracking search optimization algorithm and Integer programming approach(BackIP). Unlike, some other approaches, BackIP is a test input data generation method which includes test data generation, mutation analysis, and test suite reduction simultaneously. Experimental comparison is conducted on a widely used benchmark java programs and results show that the proposed approach achieves test data generation with mutation score up to 94% and improved test suite reduction between 70% to 94% as compared to the state-of-the-art techniques.
{"title":"BackIP: Mutation Based Test Data Generation Using Hybrid Approach","authors":"Seifu Detso Bejo, Beakal Gizachew Assefa, Sudhir Kumar Mohapatra","doi":"10.1109/ict4da53266.2021.9672216","DOIUrl":"https://doi.org/10.1109/ict4da53266.2021.9672216","url":null,"abstract":"Fault-based testing is a powerful technique to ensure the quality of software by evaluating the efficacy of the test suits and also used to check the thoroughness of testing performed by other software testing techniques. However, it is very complicated and computationally expensive testing method. Literature shows that there is a tremendous effort to give formal solutions and heuristics methods. Recently, state-of-the-art approaches based on hybrid optimization techniques have been proven to be suitable for cost effective results. This work implements and presents a multi-objective novel hybrid method by combining Backtracking search optimization algorithm and Integer programming approach(BackIP). Unlike, some other approaches, BackIP is a test input data generation method which includes test data generation, mutation analysis, and test suite reduction simultaneously. Experimental comparison is conducted on a widely used benchmark java programs and results show that the proposed approach achieves test data generation with mutation score up to 94% and improved test suite reduction between 70% to 94% as compared to the state-of-the-art techniques.","PeriodicalId":371663,"journal":{"name":"2021 International Conference on Information and Communication Technology for Development for Africa (ICT4DA)","volume":"67 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2021-11-22","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"121560245","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2021-11-22DOI: 10.1109/ict4da53266.2021.9672242
Tsion Yigzaw, Fikreselam Gared, Amare Kassaw
Photoplethysmography (PPG) signal provide advanced and simple ways for estimating heart rate (HR) information as an unremarkable system on wearable devices. In this paper, we analyze the performance of adaptive filter and machine learning (ML) algorithms for estimation of HR during physical activity. Three cascades recursive least square (RLS) and cascades normalized least mean square (NLMS) adaptive filters are developed and combined using convex combination scheme to reduce motion artifacts (MA) from the recorded PPG signal. Then, ML based spectral tracking algorithms is applied, to locate the spectral peak corresponding to HR. Four different supervised ML algorithms (Support Vector Machine, Decision Tree, K- Nearest Neighbor and Logistic Regression) are examined to track the spectral peaks and the decision tree out performs all three algorithms with an accuracy of 98.96%. Experimental results on the PPG datasets including 23 subjects used in the 2015 IEEE signal processing cup showed that the proposed approach has a very good performance by achieving an average absolute error (AAE) of 1.98 beats per minute (BPM) and the personal correlation coefficient of 0.9899. AAE result proved that the proposed method provides accurate HR estimation performance in comparison with other existing works.
{"title":"Performance Analysis of Adaptive Filter and Machine Learning Algorithms for Heart Rate Estimation Using PPG Signal","authors":"Tsion Yigzaw, Fikreselam Gared, Amare Kassaw","doi":"10.1109/ict4da53266.2021.9672242","DOIUrl":"https://doi.org/10.1109/ict4da53266.2021.9672242","url":null,"abstract":"Photoplethysmography (PPG) signal provide advanced and simple ways for estimating heart rate (HR) information as an unremarkable system on wearable devices. In this paper, we analyze the performance of adaptive filter and machine learning (ML) algorithms for estimation of HR during physical activity. Three cascades recursive least square (RLS) and cascades normalized least mean square (NLMS) adaptive filters are developed and combined using convex combination scheme to reduce motion artifacts (MA) from the recorded PPG signal. Then, ML based spectral tracking algorithms is applied, to locate the spectral peak corresponding to HR. Four different supervised ML algorithms (Support Vector Machine, Decision Tree, K- Nearest Neighbor and Logistic Regression) are examined to track the spectral peaks and the decision tree out performs all three algorithms with an accuracy of 98.96%. Experimental results on the PPG datasets including 23 subjects used in the 2015 IEEE signal processing cup showed that the proposed approach has a very good performance by achieving an average absolute error (AAE) of 1.98 beats per minute (BPM) and the personal correlation coefficient of 0.9899. AAE result proved that the proposed method provides accurate HR estimation performance in comparison with other existing works.","PeriodicalId":371663,"journal":{"name":"2021 International Conference on Information and Communication Technology for Development for Africa (ICT4DA)","volume":"94 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2021-11-22","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"134301505","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2021-11-22DOI: 10.1109/ict4da53266.2021.9672220
Wubie Belay, M. Meshesha, Dagnachew Melesew
Natural Language processing is dealing with natural language understandings and natural language generation which enable computers to understand and analyze human language. Cross-lingual Textual Entailment (CLTE) is one of the applications of NLU if there exists premise (P) as a source language and hypothesis (H) as a target language. CLTE is challenging for transferring information between under resource (Amharic) language and high resource (English) language. To solve this problem, we have proposed Cross-lingual Textual Entailment model using deep neural network approaches. We have used Bi-LSTM to transfer sequential information, XLNet for handling a position of word and its boundary, MLP for classification and prediction outputs, and FastText to word representations. Neural machine translation is utilized for translating English sentences into Amharic sentences with IBM5 alignment. We have combined Amharic dataset with SNLI dataset and annotated based on multi-way classification. The NMT predicts 96.01% of the testing accuracy. We have obtained 89.92% training and 86.89% testing accuracy for the proposed model. The issue with this research is that it ignores multiple inferences.
{"title":"Cross-lingual textual entailment using deep learning approach","authors":"Wubie Belay, M. Meshesha, Dagnachew Melesew","doi":"10.1109/ict4da53266.2021.9672220","DOIUrl":"https://doi.org/10.1109/ict4da53266.2021.9672220","url":null,"abstract":"Natural Language processing is dealing with natural language understandings and natural language generation which enable computers to understand and analyze human language. Cross-lingual Textual Entailment (CLTE) is one of the applications of NLU if there exists premise (P) as a source language and hypothesis (H) as a target language. CLTE is challenging for transferring information between under resource (Amharic) language and high resource (English) language. To solve this problem, we have proposed Cross-lingual Textual Entailment model using deep neural network approaches. We have used Bi-LSTM to transfer sequential information, XLNet for handling a position of word and its boundary, MLP for classification and prediction outputs, and FastText to word representations. Neural machine translation is utilized for translating English sentences into Amharic sentences with IBM5 alignment. We have combined Amharic dataset with SNLI dataset and annotated based on multi-way classification. The NMT predicts 96.01% of the testing accuracy. We have obtained 89.92% training and 86.89% testing accuracy for the proposed model. The issue with this research is that it ignores multiple inferences.","PeriodicalId":371663,"journal":{"name":"2021 International Conference on Information and Communication Technology for Development for Africa (ICT4DA)","volume":"61 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2021-11-22","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"127869527","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2021-11-22DOI: 10.1109/ict4da53266.2021.9671271
Tizita Dereje, Tesfamariam M Abuhay, Adane Letta, Melaku Alelign
Ethiopia, one of the Sub-Saharan countries, has been affected by preventable and treatable causes of childhood mortality. According to the Ethiopia Mini Demographic and Health Survey (EMDHS) 2019, the child mortality rate, which measures under-five child deaths per one thousand children, was 43 during the 5 years preceding the survey. This study, hence, aims to investigate risk factors and predict neonatal and infant mortality based on maternal data. To this end, data was collected from the Ethiopia Demographic and Health Surveys (EDHS) and several experiments were conducted using homogenous ensemble methods to develop a model that best identifies risk factors and predicts neonatal and infant mortality in Ethiopia. A decision tree with bagging and AdaBoost achieved an accuracy of 94.34% and 94.79% and area under ROC of 86% and 87% respectively. Naïve Bayes achieved 87.60% and 89.5% with bagging and AdaBoost. A decision tree with AdaBoost ensemble method performed better with 97.19% and 99.92% F-measure and recall, respectively. A maximum increase of 4 % accuracy for weak classifiers was achieved with the ensemble classification. As the finding suggest the interventions towards neonatal and infant mortality may need to take the factors related to maternal determinants into account. The application of heterogeneous ensemble methods is similar challenges may enhance the performance of the prediction model.
{"title":"Investigate Risk Factors and Predict Neonatal and Infant Mortality Based on Maternal Determinants using Homogenous Ensemble Methods","authors":"Tizita Dereje, Tesfamariam M Abuhay, Adane Letta, Melaku Alelign","doi":"10.1109/ict4da53266.2021.9671271","DOIUrl":"https://doi.org/10.1109/ict4da53266.2021.9671271","url":null,"abstract":"Ethiopia, one of the Sub-Saharan countries, has been affected by preventable and treatable causes of childhood mortality. According to the Ethiopia Mini Demographic and Health Survey (EMDHS) 2019, the child mortality rate, which measures under-five child deaths per one thousand children, was 43 during the 5 years preceding the survey. This study, hence, aims to investigate risk factors and predict neonatal and infant mortality based on maternal data. To this end, data was collected from the Ethiopia Demographic and Health Surveys (EDHS) and several experiments were conducted using homogenous ensemble methods to develop a model that best identifies risk factors and predicts neonatal and infant mortality in Ethiopia. A decision tree with bagging and AdaBoost achieved an accuracy of 94.34% and 94.79% and area under ROC of 86% and 87% respectively. Naïve Bayes achieved 87.60% and 89.5% with bagging and AdaBoost. A decision tree with AdaBoost ensemble method performed better with 97.19% and 99.92% F-measure and recall, respectively. A maximum increase of 4 % accuracy for weak classifiers was achieved with the ensemble classification. As the finding suggest the interventions towards neonatal and infant mortality may need to take the factors related to maternal determinants into account. The application of heterogeneous ensemble methods is similar challenges may enhance the performance of the prediction model.","PeriodicalId":371663,"journal":{"name":"2021 International Conference on Information and Communication Technology for Development for Africa (ICT4DA)","volume":"24 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2021-11-22","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"116648066","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
京公网安备 11010802042870号
京ICP备2023020795号-1
扫码关注我们
求助内容：
应助结果提醒方式：