Search

Multi-target regression (MTR) regards predictive problems with multiple numerical targets. To solve this, machine learning techniques can model solutions treating each target as a separated problem based only on the input features. Nonetheless, modelling inter-target correlation can improve predictive performance. When performing MTR tasks using the statistical dependencies of targets, several approaches put aside the evaluation of each pair-wise correlation between those targets, which may differ for each problem. Besides that, one of the main drawbacks of the current leading MTR method is its high memory cost. In this paper, we propose a novel MTR method called Multi-output Tree Chaining (MOTC) to overcome the mentioned disadvantages. Our method provides an interpretative internal tree-based structure which represents the relationships between targets denominated Chaining Trees (CT). Different from the current techniques, we compute the outputs dependencies, one-by-one, based on the Random Forest importance metric. Furthermore, we proposed a memory friendly approach which reduces the number of required regression models when compared to a leading method, reducing computational cost. We compared the proposed algorithm against three MTR methods (Single-target - ST; Multi-Target Regressor Stacking - MTRS; and Ensemble of Regressor Chains - ERC) on 18 benchmark datasets with two base regression algorithms (Random Forest and Support Vector Regression). The obtained results show that our method is superior to the ST approach regarding predictive performance, whereas, having no significant difference from ERC and MTRS. Moreover, the interpretative tree-based structures built by MOTC pose as great insight on the relationships among targets. Lastly, the proposed solution used significantly less memory than ERC being very similar in predictive performance. [ABSTRACT FROM AUTHOR]

Compromising legitimate accounts has been the most used strategy to spread malicious content on OSN (Online Social Network). To address this problem, we propose a pure text mining approach to check if an account has been compromised based on its posts content. In the first step, the proposed approach extracts the writing style from the user account. The second step comprehends the k-Nearest Neighbors algorithm (k-NN) to evaluate the post content and identify the user. Finally, Baseline Updating (third step) consists of a continuous updating of the user baseline to support the current trends and seasonality issues of user's posts. Experiments were carried out using a dataset from Twitter composed by tweets of 1000 users. All the three steps were individually evaluated, and the results show that the developed method is stable and can detect the compromised accounts. An important observation is the Baseline Updating contribution, which leads to an enhancement of accuracy superior of 60 %. Regarding average accuracy, the developed method achieved results over 93 %. [ABSTRACT FROM AUTHOR]

Contrast enhancement algorithms have been evolved through last decades to meet the requirement of its objectives. Actually, there are two main objectives while enhancing the contrast of an image: (i) improve its appearance for visual interpretation and (ii) facilitate/increase the performance of subsequent tasks (e.g., image analysis, object detection, and image segmentation). Most of the contrast enhancement techniques are based on histogram modifications, which can be performed globally or locally. The Contrast Limited Adaptive Histogram Equalization (CLAHE) is a method which can overcome the limitations of global approaches by performing local contrast enhancement. However, this method relies on two essential hyperparameters: the number of tiles and the clip limit. An improper hyperparameter selection may heavily decrease the image quality toward its degradation. Considering the lack of methods to efficiently determine these hyperparameters, this article presents a learning-based hyperparameter selection method for the CLAHE technique. The proposed supervised method was built and evaluated using contrast distortions from well-known image quality assessment datasets. Also, we introduce a more challenging dataset containing over 6200 images with a large range of contrast and intensity variations. The results show the efficiency of the proposed approach in predicting CLAHE hyperparameters with up to 0.014 RMSE and 0.935 R2 values. Also, our method overcomes both experimented baselines by enhancing image contrast while keeping its natural aspect. [ABSTRACT FROM AUTHOR]