Terms Of F-measure Research Articles

Tensor-Based Viterbi Algorithms for Collaborative Cloud-Edge Cyber-Physical-Social Activity Prediction

With the rapid development and application of smart city, Cyber-Physical-Social Systems (CPSS) as its superset is becoming increasingly important, and attracts extensive attentions. For satisfying the smart requirements of CPSS design, a cloud-edge collaborative CPSS framework is first proposed in this paper. Then Coupled-Hidden-Markov-Model (CHMM) and tensor algebra are used to improve existing activity prediction methods for providing CPSS with more intelligent decision support. There are three key features (timing, periodicity and correlation) implied in CPSS data from multi-edge, which affects the accuracy of activity prediction. Thus, these features are synthetically integrated into improved Tensor-based CHMMs (T-CHMMs) to enhance the prediction accuracy. Based on the multi-edge CPSS data, three Tensor-based Viterbi Algorithms (TVA) are correspondingly proposed to solve the prediction problem for T-CHMMs. Compared with traditional matrix-based methods, the proposed TVA could more accurately compute the optimal hidden state sequences under given observation sequences. Finally, the comprehensive performances of proposed models and algorithms are validated on three open datasets by self-comparison and other-comparison. The experimental results show that the proposed methods is superior to the compared three classical methods in terms of F1 measure, average precision and average recall.

Transfer Learning based Low Shot Classifier for Software Defect Prediction

Background: The rapid growth and increasing complexity of software applications are causing challenges in maintaining software quality within constraints of time and resources. This challenge led to the emergence of a new field of study known as Software Defect Prediction (SDP), which focuses on predicting future defect in advance, thereby reducing costs and improving productivity in software industry. Objective: This study aimed to address data distribution disparities when applying transfer learning in multi-project scenarios, and to mitigate performance issues resulting from data scarcity in SDP. Methods: The proposed approach, namely Transfer Learning based Low Shot Classifier (TLLSC), combined transfer learning and low shot learning approaches to create an SDP model. This model was designed for application in both new projects and those with minimal historical defect data. Results: Experiments were conducted using standard datasets from projects within the National Aeronautics and Space Administration (NASA) and Software Research Laboratory (SOFTLAB) repository. TLLSC showed an average increase in F1-Measure of 31.22%, 27.66%, and 27.54% for project AR3, AR4, and AR5, respectively. These results surpassed those from Transfer Component Analysis (TCA+), Canonical Correlation Analysis (CCA+), and Kernel Canonical Correlation Analysis plus (KCCA+). Conclusion: The results of the comparison between TLLSC and state-of-the-art algorithms, namely TCA+, CCA+, and KCCA+ from the existing literature consistently showed that TLLSC performed better in terms of F1-Measure. Keywords: Just-in-time, Defect Prediction, Deep Learning, Transfer Learning, Low Shot Learning

Manifold-based sparse representation for opinion mining

What the consumer thinks about an organization's products, services, and events is a crucial performance indicator for businesses. The brief opinion pieces were quickly published on websites and social media platforms and have been analyzed by machine learning methods. The classical text feature representation methods suffer from high dimensionality, sparsity, noisy, irrelevant and redundant information. This paper focuses on how to enhance feature representation for opinion mining. Some nonlinear feature selection methods based on manifold assumption have been exploited to resolve these problems. The inherent manifold configuration was commonly ascertained through a nearest neighbor graph, whereby the neighbors in the current techniques may exhibit diverse polarities. To alleviate this burden, it is proposed to exploit both manifold assumption and sparse property as prior knowledge for opinion representation to learn intrinsic structure from data. First, the graph representation of user reviews based on the mentioned prior knowledge is learned. Then, the spectral properties of the learned graph are exploited to present data in a new feature space. The proposed algorithm is applied to four various common input features on two benchmark datasets, the Internet Movie Database (IMDB) and the Amazon review dataset. Our experiments reveal that the proposed algorithm yields considerable enhancements in terms of F-measure, accuracy, and other standard performance measures compared to the combination of state-of-the-art features with various classifiers. The highest classification accuracies of 99.15 and 91.97 are obtained in the proposed method on IMDB and Amazon using a linear SVM classifier, respectively. The impact of the parameters of the proposed algorithm is also investigated in this paper. The incorporation of a sparse manifold-based representation has led to noteworthy advancements beyond the baseline, and this success serves to validate the underlying assumptions.

Extractive Summarization of Telugu Text Using Modified Text Rank and Maximum Marginal Relevance

With the rapid growth of digital content, there is a need for an automatic text summarizer to provide short text from a long text document. Many research works have been presented for extractive text summarization (ETS). This article mainly focuses on the graph-based ETS approach for multiple Telugu text documents. A modified Text-Rank algorithm is employed with the noun and verb count of each sentence in the text as the initial score of each node. To get the optimal features, a novel feature selection algorithm called improved Flamingo Search Algorithm is proposed in this article. Though graph-based ETS is an important approach, the generated summaries are redundant. To reduce the redundancy in the generated summary, maximum marginal relevance is combined with the modified Text-Rank. Different word-embedding techniques such as Fast-Text, Word2vec, TF-IDF, and one-hot encoding are utilized to experiment with the proposed approach. The performance of the proposed text summarization approach is evaluated with BLEU and ROUGE in terms of F-measure, precision, and recall.

Injury severity prediction of cyclist crashes using random forests and random parameters logit models

Cycling provides numerous benefits to individuals and to society but the burden of road traffic injuries and fatalities is disproportionately sustained by cyclists. Without awareness of the contributory factors of cyclist death and injury, the capability to implement context-specific and appropriate measures is severely limited. In this paper, we investigated the effects of the characteristics related to the road, the environment, the vehicle involved, the driver, and the cyclist on severity of crashes involving cyclists analysing 72,363 crashes that occurred in Great Britain in the period 2016–2018. Both a machine learning method, as the Random Forest (RF), and an econometric model, as the Random Parameters Logit Model (RPLM), were implemented.Three different RF algorithms were performed, namely the traditional RF, the Weighted Subspace RF, and the Random Survival Forest. The latter demonstrated superior predictive performances both in terms of F-measure and G-mean. The main result of the Random Survival Forest is the variable importance that provides a ranked list of the predictors associated with the fatal and severe cyclist crashes. For fatal classification, 19 variables showed a normalized importance higher than 5% with the second involved vehicle manoeuvring and the gender of the driver of the second vehicle having the greatest predictive ability. For serious injury classification, 13 variables showed a normalized importance higher than 5% with the bike leaving the carriageway having the greatest normalized importance. Furthermore, each path from the root node to the leaf nodes has been retraced the way back generating 361 if-then rules with fatal crash as consequent and 349 if-then rules with serious injury crash as consequent. The RPLM showed significant unobserved heterogeneity in the data finding four normal distributed indicator variables with random parameters: cyclist age ≥ 75 (fatal prediction), cyclist gender male (fatal and serious prediction), and driver aged 55–64 (serious prediction). The model’s McFadden Pseudo R2 is equal to 0.21, indicating a very good fit. Furthermore, to understand the magnitude of the effects and the contribution of each variable to injury severity probabilities the pseudo-elasticity was assessed, gaining valuable insights into the relative importance and influence of the variables.The RF and the RPLM resulted complementary in identifying several roadways, environmental, vehicle, driver, and cyclist-related factors associated with higher crash severity. Based on the identified contributory factors, safety countermeasures useful to develop strategies for making bike a safer and more friendly form of transport were recommended.

CACO: A core-attachment method with cross-species functional ortholog information to detect human protein complexes.

Protein complexes play an essential role in living cells. Detecting protein complexes is crucial to understand protein functions and treat complex diseases. Due to high time and resource consumption of experiment approaches, many computational approaches have been proposed to detect protein complexes. However, most of them are only based on protein-protein interaction (PPI) networks, which heavily suffer from the noise in PPI networks. Therefore, we propose a novel core-attachment method, named CACO, to detect human protein complexes, by integrating the functional information from other species via protein ortholog relations. First, CACO constructs a cross-species ortholog relation matrix and transfers GO terms from other species as a reference to evaluate the confidence of PPIs. Then, a PPI filter strategy is adopted to clean the PPI network and thus a weighted clean PPI network is constructed. Finally, a new effective core-attachment algorithm is proposed to detect protein complexes from the weighted PPI network. Compared to other thirteen state-of-the-art methods, CACO outperforms all of them in terms of F-measure and Composite Score, showing that integrating ortholog information and the proposed core-attachment algorithm are effective in detecting protein complexes. CACO is available at https://csuligroup.com/CACO/ and https://github.com/CSUBioGroup/CACO.

A Network Clustering Algorithm for Protein Complex Detection Fused with Power-Law Distribution Characteristic

Network clustering for mining protein complexes from protein–protein interaction (PPI) networks has emerged as a prominent research area in data mining and bioinformatics. Accurately identifying complexes plays a crucial role in comprehending cellular organization and functionality. Network characteristics are often useful in enhancing the performance of protein complex detection methods. Many protein complex detection algorithms have been proposed, primarily focusing on local micro-topological structure metrics while overlooking the potential power-law distribution characteristic of community sizes at the macro global level. The effective use of this distribution characteristic information may be beneficial for mining protein complexes. This paper proposes a network clustering algorithm for protein complex detection fused with power-law distribution characteristic. The clustering algorithm constructs a cluster generation model based on scale-free power-law distribution to generate a cluster with a dense center and relatively sparse periphery. Following the cluster generation model, a candidate cluster is obtained. From a global perspective, the number distribution of clusters of varying sizes is taken into account. If the candidate cluster aligns with the constraints defined by the power-law distribution function of community sizes, it is designated as the final cluster; otherwise, it is discarded. To assess the prediction performance of the proposed algorithm, the gold standard complex sets CYC2008 and MIPS are employed as benchmarks. The algorithm is compared to DPClus, IPCA, SEGC, Core, SR-MCL, and ELF-DPC in terms of F-measure and Accuracy on several widely used protein–protein interaction networks. The experimental results show that the algorithm can effectively detect protein complexes and is superior to other comparative algorithms. This study further enriches the connection between analyzing complex network topology features and mining network function modules, thereby significantly contributing to the improvement of protein complex detection performance.

Exploiting social trust via weighted voting strategy for recommendation systems improvement

Recently, social trust information has become a significant additional factor in obtaining high-quality recommendations. It has also helped to alleviate the problems of collaborative filtering. In this paper, we exploit explicit and implicit trust relations and incorporate them to take advantage of more ratings (as they exist) of trusted neighbors to mitigate the sparsity issue. We further apply the idea of weighted voting of the ensemble classifier for the election of the most appropriate trust neighbors’ ratings. Additionally, the certainty of these elected rating values was confirmed by calculating their reliability using a modified version of Pearson’s Correlation Coefficient. Finally, we applied the K-Nearest Neighbors method with a linear combination of original and trust-elected ratings using a contribution weight to obtain the best prediction value. Extensive experiments were conducted on two real-world datasets to show that our proposed approach outperformed all comparable algorithms in terms of both coverage and accuracy. Specifically, the improvement ratio ranged approximately from 4% as a minimum to 20% on FilmTrust, and to 10% on Epinions as a maximum in terms of Fmeasure between the inverse of Mean Absolute Error (accuracy) and coverage.

Long Short-term Memory Applied on Amazon's Stock Prediction

More and more investors are paying attention to how to use data mining technology into stock investing decisions as a result of the introduction of big data and the quick expansion of financial markets. Machine learning can automatically apply complex mathematical calculations to big data repeatedly and faster. The machine model can analyze all the factors and indicators affecting stock price and achieve high efficiency. Based on the Amazon stock price published on Kaggle, this paper adopts the Long Short-term Memory (LSTM) method for model training. The Keras package in the Python program is used to normalize the data. The Sequence model in Keras establishes a two-layer LSTM network and a three-layer LSTM network to compare and analyze the fitting effect of the model on stock prices. By calculating RMSE and RMPE, the study found that the stock price prediction accuracy of two-layer LSTM is similar to that of three-layer LSTM. In terms of F-measure and Accuracy, the LSTM model of the three-layer network is significantly better than the LSTM model of the two-layer network layer. In general, the LSTM model can accurately predict stock price. Therefore, investors will know the upward or downward trend of stock prices in advance according to the prediction results of the model to make corresponding decisions.

Dynamic Data Augmentation Based on Imitating Real Scene for Lane Line Detection

With the rapid development of urban ground transportation, lane line detection is gradually becoming a major technological direction to help to realize safe vehicle navigation. However, lane line detection results may have incompleteness issues, such as blurry lane lines and disappearance of the lane lines in the distance, since the lane lines may be heavily obscured by vehicles and pedestrians on the road. In addition, low-visibility environments also pose a challenge for lane line detection. To solve the above problems, we propose a dynamic data augmentation framework based on imitating real scenes (DDA-IRS). DDA-IRS contains three data augmentation strategies that simulate different realistic scenes (i.e., shadows, dazzle, and crowded). In this way, we expand from a limited scene dataset to realistically fit multiple complex scenes. Importantly, DDA-IRS is a lightweight framework that can be integrated with a variety of training-based models without modifying the original model. We evaluate the proposed DDA-IRS on the CULane dataset, and the results show that the data-enhanced model outperforms the baseline model by 0.5% in terms of F-measure. In particular, the F-measure of the “Normal”, “Crowded”, “Shadow”, “Arrow”, and “Curve” achieve a 0.4%, 0.1%, 1.6%, 0.4%, and 1.4% improvement, respectively.

시계열 데이터의 자동기계학습을 위한 메타 모델 기반의 특징 선택 방법

Feature engineering is a key step to construct machine learning model as it determines the upper limit of model’s performance. However, designing feature engineering is generally iterative, complex and time-consuming step. Also, the large scale of time series data is rapidly generated from the industry, but there is a shortage of data scientists to handle them. So, it has become necessary to automate this process. In this paper, we aim to develop a meta model-based feature selection method that can learn about which features work best given the dataset. The proposed meta-model is a kind of warm-start that can search from the candidate features that is expected to be good without starting a new search for each data. Proposed method is compared by real time-series datasets obtained from UEA & UCR Time Series Classification Repository. Then, we show the proposed method outperforms random search in terms of F1-measure.

Covid-19 Forecasting Using CNN Approach With A Halbinomial Distribution And A Linear Decreasing Inertia Weight-Based Cat Swarm Optimization

In recent times, the COVID-19 epidemic has spread to over 170 nations. Authorities all around the world are feeling the strain of COVID-19 since the total of infected people is rising as well as they does not familiar to handle the problem. The majority of current research effort is thus being directed on the analysis of COVID-19 data within the framework of the machines learning method. Researchers looked the COVID 19 data to make predictions about who would be treated, who would die, and who would get infected in the future. This might prompt governments worldwide to develop strategies for protecting the health of the public. Previous systems rely on Long Short- Term Memory (LSTM) networks for predicting new instances of COVID-19. The LSTM network findings suggest that the pandemic might be over by June of 2020. However, LSTM may have an over-fitting issue, and it may fall short of expectations in terms of true positive. For this issue in COVID-19 forecasting, we suggest using two methods such as Cat Swarm Optimization (CSO) for reducing the inertia weight linearly and then artificial intelligence based binomial distribution is used. In this proposed study, we take the COVID-19 predicting database as an contribution and normalise it using the min-max approach. The accuracy of classification is improved with the use of the first method to choose the optimal features. In this method, inertia weight is added to the CSO optimization algorithm convergence. Death and confirmed cases are predicted for a certain time period throughout India using Convolutional Neural Network with Partial Binomial Distribution based on carefully chosen characteristics. The experimental findings validate that the suggested scheme performs better than the baseline system in terms of f-measure, recall, precision, and accuracy.

Design of Intrusion Detection System using Ensemble Learning Technique in Cloud Computing Environment

The key advantage of the cloud is that it fluidly propagates to fulfil changeable requirements and provides an environment that is repeatable and can be scaled down instantly when needed. Therefore, it is necessary to protect this cloud environment from malicious attacks such as spamming, keylogging, Denial of Service (DoS), and Distributed Denial of Service (DDoS). Among these kinds of attacks, DDoS has the capability to establish a high flood of malicious attacks on the cloud environment or Software Defined Networking (SDN) based cloud environment. Hence in this work, an ensemble based deep learning technique is proposed to detect attacks in cloud and SDN based cloud environments. Here, the ensemble model is formed by combining K-means with deep learning classifiers such as Long Short term Memory (LSTM) network, Convolutional Neural Network (CNN), Recurrent Neural Network (RNN), Gated Recurrent Unit (GRU) and Deep Neural Network (DNN). Initially, preprocessing with data cleaning and standardization is applied to the input data. Meanwhile, a random forest is implemented for extracting the minimum significant features. After that, the proposed ensemble based approach is utilized for detecting the intrusion. This approach is used to enhance the performance of the deep learning classifiers without much computational complexity. This model is trained and evaluated using two datasets as CICIDS 2018 and SDN based DDOS attack datasets. The proposed approach provides better intrusion detection performance in terms of F1 measure, precision, accuracy, and recall. By using the proposed approach, the accuracy and precision value attained is 99.685 and 0.992, respectively.

Application of machine-learning algorithms to predict calving difficulty in Holstein dairy cattle

Context An ability to predict calving difficulty could help farmers make better farm-management decisions, thereby improving dairy farm profitability and welfare. Aims This study aimed to predict calving difficulty in Iranian dairy herds using machine-learning (ML) algorithms and to evaluate sampling methods to deal with imbalanced datasets. Methods For this purpose, the history records of cows that calved between 2011 and 2021 on two commercial dairy farms were used. Using WEKA software, four commonly used ML algorithms, namely naïve Bayes, random forest, decision trees, and logistic regression, were applied to the dataset. The calving difficulty was considered as a binary trait with 0, normal or unassisted calving, and 1, difficult calving, i.e. receiving any help during parturition from farm personnel involvement to surgical intervention. The average rate of difficult calving was 18.7%, representing an imbalanced dataset. Therefore, down-sampling and cost-sensitive techniques were implemented to tackle this problem. Different models were evaluated on the basis of F-measure and the area under the curve. Key results The results showed that sampling techniques improved the predictive model (P = 0.07, and P = 0.03, for down-sampling and cost-sensitive techniques respectively). F-measure ranged from 0.387 (decision tree) to 0.426 (logistic regression) with the balanced dataset. However, when applied to the original imbalanced dataset, naïve Bayes had the best performance of up to 0.388 in terms of F-measure. Conclusions Overall, sampling techniques improved the prediction model compared with original imbalanced dataset. Although prediction models performed worse than expected (due to an imbalanced dataset, and missing values), the implementation of ML algorithms can still lead to an effective method of predicting calving difficulty. Implications This research indicated the capability of ML algorithms to predict the incidence of calving difficulty within a balanced dataset, but that more explanatory variables (e.g. genetic information) are required to improve the prediction based on an unbalanced original dataset.

Comparison of different machine learning algorithms to classify patients suspected of having sepsis infection in the intensive care unit

BackgroundSepsis is a life-threatening disease that occurs as a result of the body's response to an infection. This study aims to develop a classification model for predicting patients at risk of sepsis using clinical findings and demographic information. MethodsThe study was conducted using a MIMICIII dataset which is freely available as open-access data. The synthetic minority oversampling technique (SMOTE) was applied to address the imbalanced data problem in our dataset. Through preprocessing, the dataset was cleaned and missing values were imputed. Split validation was done by dividing the dataset into training and test data for developing classification models. Six algorithms including Gaussian Naïve Bayes (NB), decision tree (DT), random forest (RF), logistic regression (LR), KNN algorithm, and XGBoost classifier were developed. A combination of evaluation metrics was employed to evaluate the performance of the proposed models. ResultsOur dataset includes 1,552,210 entries with 44 features of critically ill patients who were admitted to the ICU. Comparing the performance of developed models using different metrics showed that the RF model had the best performance in terms of F-Measure and the area under the ROC curve. The 20 top features with high importance were determined based on the RF model. ConclusionOur analysis showed that the RF model predicted sepsis with significantly higher performance in comparison to other classification models using the MIMICIII dataset. Due to the high mortality of sepsis, these kinds of studies could be supportive to prevent the side effects of the disease and lessen the risk of mortality in hospitalized patients by providing early sepsis prediction.

Covid-19 Forecasting with Deep Learning-based Half-binomial Distribution Cat Swarm Optimization

About 170 nations have been affected by the COvid VIrus Disease-19 (COVID-19) epidemic. On governing bodies across the globe, a lot of stress is created by COVID-19 as there is a continuous rise in patient count testing positive, and they feel challenging to tackle this situation. Most researchers concentrate on COVID-19 data analysis using the machine learning paradigm in these situations. In the previous works, Long Short-Term Memory (LSTM) was used to predict future COVID-19 cases. According to LSTM network data, the outbreak is expected to finish by June 2020. However, there is a chance of an over-fitting problem in LSTM and true positive; it may not produce the required results. The COVID-19 dataset has lower accuracy and a higher error rate in the existing system. The proposed method has been introduced to overcome the above-mentioned issues. For COVID-19 prediction, a Linear Decreasing Inertia Weight-based Cat Swarm Optimization with Half Binomial Distribution based Convolutional Neural Network (LDIWCSO-HBDCNN) approach is presented. In this suggested research study, the COVID-19 predicting dataset is employed as an input, and the min-max normalization approach is employed to normalize it. Optimum features are selected using Linear Decreasing Inertia Weight-based Cat Swarm Optimization (LDIWCSO) algorithm, enhancing the accuracy of classification. The Cat Swarm Optimization (CSO) algorithm’s convergence is enhanced using inertia weight in the LDIWCSO algorithm. It is used to select the essential features using the best fitness function values. For a specified time across India, death and confirmed cases are predicted using the Half Binomial Distribution based Convolutional Neural Network (HBDCNN) technique based on selected features. As demonstrated by empirical observations, the proposed system produces significant performance in terms of f-measure, recall, precision, and accuracy.

HS-Gen: a hypersphere-constrained generation mechanism to improve synthetic minority oversampling for imbalanced classification

Mitigating the impact of class-imbalance data on classifiers is a challenging task in machine learning. SMOTE is a well-known method to tackle this task by modifying class distribution and generating synthetic instances. However, most of the SMOTE-based methods focus on the phase of data selection, while few consider the phase of data generation. This paper proposes a hypersphere-constrained generation mechanism (HS-Gen) to improve synthetic minority oversampling. Unlike linear interpolation commonly used in SMOTE-based methods, HS-Gen generates a minority instance in a hypersphere rather than on a straight line. This mechanism expands the distribution range of minority instances with significant randomness and diversity. Furthermore, HS-Gen is attached with a noise prevention strategy that adaptively shrinks the hypersphere by determining whether new instances fall into the majority class region. HS-Gen can be regarded as an oversampling optimization mechanism and flexibly embedded into the SMOTE-based methods. We conduct comparative experiments by embedding HS-Gen into the original SMOTE, Borderline-SMOTE, ADASYN, k-means SMOTE, and RSMOTE. Experimental results show that the embedded versions can generate higher quality synthetic instances than the original ones. Moreover, on these oversampled datasets, the conventional classifiers (C4.5 and Adaboost) obtain significant performance improvement in terms of F1 measure and G-mean.

An optimized fuzzy deep learning model for data classification based on NSGA-II

As a powerful paradigm, deep learning (DL) models have been used in many applications for classification tasks in images, text, and audio. Through DL models, we can learn task-driven features from big data. However, DL models are fully deterministic and cannot handle uncertain and imprecise data. DL models are often sensitive to noise in data and do not operate well in areas where data are vague. Moreover, when there is a large feature set or high-dimensional data with irrelevant and redundant features, the DL models’ performance decreases in classification due to training with irrelevant features. To gain reliable results in such high-dimensional problems, DL models require a large amount of data which usually grows exponentially concerning the number of features. Data uncertainty problems and irrelevant features in DL models cause low performance in classification tasks. This paper proposes an optimized fuzzy deep learning (OFDL) model for data classification based on Non-Dominated Sorting Genetic Algorithm II (NSGA-II). OFDL utilizes optimization in the composition of DL and fuzzy learning via the NSGA-II in multi-modal learning. To achieve effective classification, OFDL first considers intelligent feature selection by finding the best trade-offs between two conflicting objective functions, minimizing the number of features, and maximizing the accuracy (maximizing weights of selected features). Next, to reach optimized backpropagation and fuzzy membership functions, OFDL utilizes Pareto optimal solutions for multi-objective optimization using NSGA-II based on their objective functions. Furthermore, the fusion layer in OFDL fused optimized views of DL and fuzzy learning that provides a high-level representation of inputs and optimum features for classification tasks where the data contain high uncertainties and noises. This functionality gives valuable attributes during classification since identifying and selecting appropriate features ensures prompt and correct class. Also, it provides deep insight into tackling the effect of ambiguous data and each feature’s uncertainty in the classification tasks. The examination of OFDL reveals good performance in terms of F-measure, accuracy, recall, precision, and True Positive Rate (TPR) compared to fuzzy classifiers. Furthermore, OFDL has higher accuracy in classification tasks than earlier fuzzy DNN models.

A new approach for road extraction using data augmentation and semantic segmentation

Accurate road extraction from remote sensing images is a challenging task. Several methods of extraction have been developed but the precision of extraction is still limited for the unpaved and small-width roads. This paper proposes an accurate road extraction approach called DAA-SSEG since it uses data augmentation architecture (DAA) and semantic segmentation model (SSEG). The proposed approach DAA-SSEG is based on a modified full convolutional neural network that overcomes the vanishing gradient and the training saturation issues. It recognizes roads at the pixel level. Furthermore, The DAA-SSEG approach uses a new plan of data augmentation based on geometric transformation and images refinement techniques. It allows getting a richer dataset thus better training and an accurate extraction. The experiment denotes that the proposed approach DAA-SSEG, that combine data augmentation architecture and semantic segmentation method, outperforms some state-of-the-art methods in terms of F-measures. The results demonstrate that it ensures accurate extraction of unpaved and small-width roads, in urban and rural areas. Moreover, the proposed approach distinguishes between roads and trails and can extract some roads not labeled beforehand.

A deep learning-based expert finding method to retrieve agile software teams from CQAs

Currently, many software companies are looking to assemble a team of experts who can collaboratively carry out an assigned project in an agile manner. The most ideal members for an agile team are T-shaped experts, who not only have expertise in one skill-area but also have general knowledge in a number of related skill-areas. Existing related methods have only used some heuristic non-machine learning models to form an agile team from candidates, while machine learning has been successful in similar tasks. In addition, they have only used the number of candidates’ documents in various skill-areas as a resource to estimate the candidates’ T-shaped knowledge to work in an agile team, while the content of their documents is also very important. To this end, we propose a multi-step method that rectifies the drawbacks mentioned. In this method, we first pick out the best possible candidates using a state-of-the-art model, then we re-estimate their relevant knowledge for working in the team with the help of a deep learning model, which uses the content of the candidates’ posts on StackOverflow. Finally, we select the best possible members for the given agile team from among these candidates using an integer linear programming model. We perform our experiments on two large datasets C# and Java, which comprise 2,217,366 and 2,320,883 posts from StackOverflow, respectively. On datasets C# and Java, our method selects, respectively, 68.6% and 55.2% of the agile team members from among T-shaped experts, while the best baseline method only selects, respectively, 49.1% and 40.2% of the agile team members from among T-shaped experts. In addition, the results show that our method outperforms the best baseline method by 8.1% and 11.4% in terms of F-measure on datasets C# and Java, respectively.