The university student's dropout is a problem that affects the governments, institutions and students. It has negative effects on the high expenditure in the administrative and academic resources. Predicting dropout has become an advantage for university administrators because it allows discovering students that are at risk of dropout as well as develop actions that allow taking decisions in a timely manner. This research presents a neural network approach through the application of multilayer perceptrom algorithms and radial basis function. As input variables to the models, 11 factors were considered, which produce a negative influence in the desertion at the universities; the data was obtained from a survey of 2670 students of a Public University in Ecuador. The results showed that there is no significant difference in the accuracy rates of the proposed models which correspond to 96.3% for multilayer perceptrom and 96.8% for radial basis function. As a conclusion, the studied models could be considered as an optimal option in terms of accuracy and concordance to predict dropout at the universities.

With more and more frequency, users communicate with each other on social media. Many users start on Twitter or Facebook to find friends who have the same hobby. Our study proposes a method to estimate the users’ interests (hobby) based on tweets on Twitter. One tweet does not, in and of itself, contain a lot of information, and some tweets are not related to the user’s hobby. Therefore, we propose a reliable hobby estimation method by extracting features from multiple, sequential tweets. The proposed method uses Recurrent Neural Networks (RNN) which can accommodate time-series information. We also used a Convolutional Neural Networks (CNN) which can treat contextual information. We used an averaged vector of word distributed representation as a feature. Using the proposed method based on Long Short-Term Memory Recurrent Neural Networks (LSTM-RNN), we obtained a 23.72% improvement as compared with a baseline method using a Random Forest (RF) regression as a machine learning algorithm.

Recent advancements in deep learning (DL) frameworks based on deep neural networks (DNN) drastically improved accuracy in image recognition, natural language processing and other applications.The key advantage of DL is systematic approach for independent training of groups of DNN layers including unsupervised training of auto-encoders for hierarchical representation of raw input data (i.e., automatic feature selection and dimensionality reduction) and supervised re-training of several final layers in the transfer learning that compensate for data incompleteness.However, severe data limitations and/or absence of relevant problem for transfer learning can drastically reduce advantages of DNN-based DL.For example, pure data-driven auto-encoders dealing with high-dimensional input data require large amount of data for effective operation.However, hierarchical data representations can be also implemented without NN.Previously we have shown robustness of boosting-like algorithms for effective utilization of existing domain knowledge (e.g.analytical models) via discovery of compact ensembles of complementary low-complexity components.This approach can tolerate significant data incompleteness and boost accuracy of individual base models as was demonstrated in cardiac diagnostics applications.Here we argue that hybrid DL framework with auto-encoders replaced by components discovered by boosting followed by supervised NN could be more tolerant to data incompleteness compared to pure DNN-based DL.Illustrations based on cardio data from www.physionet.org are presented.The proposed framework could be utilized in many applications dealing with incomplete data including personalized medicine and rare or complex abnormalities.

This documentary research purposed to design the System Framework for an intelligent question bank and examination system.It was divided into three periods: (1) the synthesis for functions in the question bank system; (2) design of the system framework for an intelligent question bank and examination system; (3) to evaluate the appropriateness of the system framework for an intelligent question bank and examination system.A content analysis was applied to analyze the data.The results found that:1.The question bank system consisted of five modules: 1. User Management, 2. Question Management, 3. Examination Management, 4. Evaluation Management and 5. Scoring Management.2. The system framework of the question bank system consisted of three parts: 1. Relevant persons for the question bank system: teachers, students, and administrators; 2. Framework of the question bank system consisted of five main modules and 15 sub-modules: 1. User Management: 1.1 Login, 1.2 Edit Profiles, 1.3 Set Permission. 2.

One of the most important aspects of financial risk is credit risk management. Effective credit rating models are crucial for the credit institution in assessing credit applications, they have been widely studied in the field of statistics and machine learning. Given that small improvements in credit rating systems can generate significant profits, any improvement is of high interest to banks and financial institutions. The ensemble methods are a set of algorithms whose individual decisions are combined to perform classification tasks. In this work, we propose an enhanced experimental comparative study of five ensemble methods associated with seven base classifiers using six public credit scoring datasets. Four popular evaluation metrics, including area under the curve (AUC), accuracy, false positive rate (FPR) and Time taken to build the model, are employed to measure the performance of models. The experimental results and statistical tests show that Pegasos model has a better overall performance than the other methods analyzed her for Boosting and Credal Decision Tree (CDT) model has a better overall performance than the other algorithms in the case of Bagging, Random Subspace, DECORATE and Rotation Forest.

The exponential growth in the number of complex datasets every year requires\nmore enhancement in machine learning methods to provide robust and accurate\ndata classification. Lately, deep learning approaches have achieved surpassing\nresults in comparison to previous machine learning algorithms. However, finding\nthe suitable structure for these models has been a challenge for researchers.\nThis paper introduces Random Multimodel Deep Learning (RMDL): a new ensemble,\ndeep learning approach for classification. RMDL solves the problem of finding\nthe best deep learning structure and architecture while simultaneously\nimproving robustness and accuracy through ensembles of deep learning\narchitectures. In short, RMDL trains multiple randomly generated models of Deep\nNeural Network (DNN), Convolutional Neural Network (CNN) and Recurrent Neural\nNetwork (RNN) in parallel and combines their results to produce better result\nof any of those models individually. In this paper, we describe RMDL model and\ncompare the results for image and text classification as well as face\nrecognition. We used MNIST and CIFAR-10 datasets as ground truth datasets for\nimage classification and WOS, Reuters, IMDB, and 20newsgroup datasets for text\nclassification. Lastly, we used ORL dataset to compare the model performance on\nface recognition task.\n

In this paper, we present a novel system to analyze human body motions (actions) for recognizing human actions by using 3D videos (RGB and depth data). We apply the Bag-of-Features techniques for recognizing human actions by extracting local-spatial temporal features from all video frames. Feature vectors are computed in two steps: The first step consists of detecting all interest keypoints from RGB video frames by using Speed-Up Robust Features detector; then the motion points are filtered by using Motion History Image and Optical Flow, and these important motion points are aligned to the depth frame sequences. In the second step, the feature vectors are computed by using a Histogram of Orientation Gradient descriptor, this descriptor is applied around these motion points from both RGB and depth channels, then the feature vector values are combined in one RGBD feature vector. Finally, the k-means clustering and multi-class Support Vector Machines are used for the action classification task. Our system is invariant to scale, rotation and illumination. All tested results are computed from a dataset that is available to the public and often used in the community. This new features combination method is help to reach recognition rates superior to other publications on the dataset.

Abstract-Disaster preparedness and disaster management plan is one of the main concerns of the government to reduce the damage in human lives and economic loss. As a response, the researchers decided to develop a data-driven architectural framework for LGUs in Disaster Preparedness and Management System. The researchers used descriptive research method to obtain data and information in identifying the attributes needed in employing disaster preparedness and management plan for data mining processes. Qualitative research method was also used to develop the architectural framework to be designed from the identified attributes in disaster preparedness and management plan and to test the viability of the designed architectural framework in disaster preparedness and management system. Respondents of this study are the Municipal Disaster Risk Reduction Management Council (MDRRMC), Department of Agriculture (DA), Municipal Engineering Office (MEO), Municipal Social Welfare and Development (MSWD), and the Local Government Units (LGUs). Based on the assessment of the user (highest scale is 5), the developed architectural framework has weighted mean of 4.58 in terms of efficiency, effectiveness and impact which means that this will serve as a significant tool for decision making to help the LGUs to have a concrete, effective and efficient disaster preparedness and management plan based on their needs to reduce vulnerability in human lives, infrastructure and agriculture.

The multi-objective problem is particularly difficult in practical engineering applications, so more and more scholars have studied the problem to find the true Pareto optimal solution. In order to improve the convergence performance of multi-objective optimization algorithm and diversity, this paper proposes a multi-objective optimization algorithm based on chaos particle swarm optimization algorithm: using Logistic mapping sequences in solution in the particle swarm algorithm is updated; introducing the crossover operator of normal distribution to improve the diversity of the population; and using simplified mesh reduction and gene exchange to improve the performance of the algorithm. Compared with the MOPSO, NSGA-II and MOEA/D algorithms, it is shown that the proposed algorithm has good performance and can effectively solve the multi-objective optimization problem.

Problem of transferring and testing self-driving algorithms developed in virtual environment to a physical environment is explored by transferring a Convolutional Neural Network based self-driving car steering algorithm from virtual environment to physical RC card based environment for validation and testing as a step on the way for full scale selfdriving car tests. In the process a novel approach for synthetic training data generation from single camera is developed, thus reducing the real world physical requirements for the algorithm and demonstrating the improved self-driving algorithm development pipeline from fully virtual environments to scaled physical models, to full self-driving cars, potentially leveraging the global developer community for development.