Narrow Results

Scene recognition is an important computer vision task that has evolved from the study of the biological visual system. Its applications range from video surveillance, autopilot systems, to robotics. The early works were based on feature engineering that involved the computation and aggregation of global and local image descriptors. Several popular image features such as SIFT, SURF, HOG, ORB, LBP, KAZE, etc. have been proposed and applied to the task with successful results.

Features can be either computed from the entire image on a global scale, or extracted from local sub-regions and aggregated across the image. Suitable classifier models are deployed that learn to classify these features. This review paper analyzes several of these handcrafted features that have been applied to the scene recognition task over the past decades, and tracks the transition from the traditional feature engineering to deep learning which forms the current state of the art in computer vision. Deep learning is now deemed to have overtaken feature engineering in several computer vision applications. Deep convolutional neural networks and vision transformers are the current state of the art for object recognition. However, scenes from urban landscapes are bound to contain similar objects posing a challenge to deep learning solutions for scene recognition. In our study, a critical analysis of feature engineering and deep learning methodologies for scene recognition is provided, and results on benchmark scene datasets are presented, concluding with a discussion on challenges and possible solutions that may facilitate more accurate scene recognition.

Human emotion classification without bias and unfairness is challenging because most existing image-based methods are directly or indirectly affected by subjectivity. Therefore, we propose an EEG (Electroencephalogram) based model for an accurate emotion classification without the effect of subjectivity. The captured EEG signals are converted into Delta, Theta, Alpha, Beta, and Gama frequency bands. As emotions change, the frequency bands change and provide unique patterns for each emotion irrespective of different persons. With this observation, the statical features, namely, mean, standard deviation, variance, and kurtosis, and frequency-based features, namely, Power Spectral Density (PSD) and Petrosian Fractal Dimension (PFD) are extracted. To integrate the strength of spatial and frequency-based features, the features are supplied to quadratic discriminative analysis for the final classification. The experiments on the benchmark datasets, DEAP and SEED-IV, achieve 99.40% and 91.97% accuracy, respectively. A comparison with state-of-the-art methods shows that the method performs very well on some datasets.

Traffic forecasting is an integral part of modern intelligent transportation systems. Although many techniques have been proposed in the literature to address the problem, most of them focus almost exclusively on forecasting accuracy and ignore other important aspects of the problem. In the paper at hand, a new method for both accurate and fast large-scale traffic forecasting, named “sparse feature regression”, is presented. Initially, a set of carefully selected features is extracted from the available traffic data. Then, some of the initial features are sparsified, namely they are transformed into sets of sparse features. Finally, a linear regression model is designed using the sparse feature set, which is trained by solving an optimization problem using a sparse approximate pseudoinverse as a preconditioner. We evaluated the proposed method by conducting experiments on two real-world traffic datasets, and the experimental results showed that the method presents the best balance between accuracy of predictions and time required for achieving them, in comparison with a set of benchmark models.

Text classification involves organizing textual information into predefined classes, a task which is particularly useful in domains like sentiment analysis, spam detection, and content labeling. In India, where a massive amount of information is generated daily through newspapers and social media, Hindi is one of the most widely used and spoken languages. However, there is limited research on Hindi text classification and, particularly, regarding Hindi news classification. This paper presents a research study to classify Hindi news articles published in Hindi-language newspapers in India by using and comparing various Machine Learning (ML) and Deep Learning (DL) algorithms. To prepare the textual news data for classification, pre-processing and feature engineering techniques, such as count vectorizer, Tf-Idf vectorizer and Doc2Vec, were used and applied to convert texts into vectors. This pre-processing step on the textual data was very challenging due to the presence of multimodal words, conjunctions, punctuation, and special characters in Hindi texts. The study considered Hindi news headlines from predetermined categories (Science, Sports, Entertainment and Business) and, among the different ML and DL models tested and evaluated, Linear Regression with Doc2Vec vectorizer and SGD classifier with Tf-Idf vectorizer produced best accuracies of 97.04% and 96.59%, respectively. The best performing DL model was found to be the Bi-LSTM with an accuracy of approximately 97% on the testing data.

This paper performed some exploratory data visualization on this data set. The nature and representation of input data was found out and the preliminary feature selection was conducted in this step. And this paper performed data preprocessing and feature engineering on this data set, which had critical importance of the accuracy of prediction results. The paper built multiple regression models on the missing values prediction in the testing set. The paper implemented various data mining algorithms to build predictive models, including Gaussian Naive Bayes classifier, K-Nearest Neighbors (K-NN) algorithm, Multi-layer Perceptron (MLP), Logistic regression, random forest and XGBoost. After the experiments, XGBoost classifier could give the best result among all the models.

Modern biotechnologies have generated huge amount of OMIC data, among which transcriptomes and methylomes are two major OMIC types. Transcriptomes measure the expression levels of all the transcripts while methylomes depict the cytosine methylation levels across a genome. Both OMIC data types could be generated by array or sequencing. And some studies deliver many more features (the number of features is denoted as $p$) for a sample than the number $n$ of samples in a cohort, which induce the “large $p$ small $n$” paradigm. This study focused on the classification problem about OMIC with “large $p$ small $n$” paradigm. A Siamese convolutional network was utilized to transform the OMIC features into a new space with minimized intra-class distances and maximized inter-class distances between the samples. The proposed feature engineering algorithm SiaCo was comprehensively evaluated using both transcriptome and methylome datasets. The experimental data showed that SiaCo generated SiaCo features with improved classification accuracies for binary classification problems, and achieved improvements on the independent test dataset. The individual SiaCo features did not show better inter-class discrimination powers than the original OMIC features. This may be due to that the Siamese convolutional network optimized the collective performances of the SiaCo features, instead of the individual feature’s discrimination power. The inherent transformation nature of the Siamese twin network also makes the SiaCo features lack of interpretability. The source code of SiaCo is freely available at http://www.healthinformaticslab.org/supp/resources.php.

The Large Synoptic Survey Telescope will begin its survey in 2022 and produce terabytes of imaging data each night. To work with this massive onset of data, automated algorithms to classify astronomical light curves are crucial. Here, we present a method for automated classification of photometric light curves for a range of astronomical objects. Our approach is based on the gradient boosting of decision trees, feature extraction and selection, and augmentation. The solution was developed in the context of The Photometric LSST Astronomical Time Series Classification Challenge (PLAsTiCC) and achieved one of the top results in the challenge.

As an important tool to measure the current situation of the whole stock market, the stock index has always been the focus of researchers, especially for its prediction. This paper uses trend types, which are received by clustering price series under multiple time scale, combined with the day-of-the-week effect to construct a categorical feature combination. Based on the historical data of six kinds of Chinese stock indexes, the CatBoost model is used for training and predicting. Experimental results show that the out-of-sample prediction accuracy is 0.55, and the long–short trading strategy can obtain average annualized return of 34.43%, which is a great improvement compared with other classical classification algorithms. Under the rolling back-testing, the model can always obtain stable returns in each period of time from 2012 to 2020. Among them, the SSESC’s long–short strategy has the best performance with an annualized return of 40.85% and a sharp ratio of 1.53. Therefore, the trend information on multiple time-scale features based on feature engineering can be learned by the CatBoost model well, which has a guiding effect on predicting stock index trends.

Motivated by the rise of social networks, as well as their applications in fields such as genetics or marketing, the node classification problem is one of the most actively researched topics within network-based machine learning. In contrast to methodologies based on neural networks or relational probabilistic algorithms, the main objective of this chapter is to provide an alternative by using classical algorithms (such as decision trees). This chapter aims to illustrate how, by using methodologies inherent to community detection, datasets can be enriched in such a way that the performance of these algorithms improves. A new measure of homophily may be proposed based on that improvement ratio.

Incremental software development replaces monolithic-type development by offering a series of releases with additive functionality. To create optimal value under existing project constraints, the question is what should be done when? Release planning is giving the answer. It determines proper priorities and assigns features to releases. Comprehensive stakeholder involvement ensures a high degree of applicability of the results. The formal procedure of release planning is able to consider different criteria (urgency, importance) and to bring them together in a balanced way. Release planning is based on (estimates of) the implementation effort. In addition, constraints related to risk, individual resources necessary to implement the proposed features, money, or technological dependencies can be easily adopted into the release planning approach presented in this article.

Releases are known to be new versions of an evolving product. However, the idea of a release is not restricted to this, but can be applied to any type of periodic development where a release would correspond to an annual or quarterly time period. The special case of one release called prioritization is of even larger applicability wherever competing items have been selected under additional constraints.

An informal and later a formal problem description of the release planning problem is given. Ad hoc or just experience-based planning techniques are not able to accommodate size, complexity and the high degree of uncertainty of the problem. Plans generated in this way will typically result in unsatisfied customers, time and budget overruns, and a loss in market share. As a consequence of the analysis of the current state-of-the practice, we propose a more advanced approach based on the strengths of intelligent software engineering decision support.

Existing release planning methods and tool support are analyzed. An intelligent tool support called ReleasePlanner® is presented. The web-based tool is based on an iterative and evolutionary solution procedure and combines the computational strength of specialized optimization algorithms with the flexibility of intelligent decision support. It helps to generate and evaluate candidate solutions. As a final result, a small number of most promising alternative release plans are offered to the actual decision-maker. Special emphasis is on facilitating what-if scenarios and on supporting re-planning. Different usage scenarios and a case study project are presented. Practical experience from industrial application of ReleasePlanner® is included as well. Future directions of research are discussed.