Search

This study addresses a Biomass data classification problem. One of the objectives is to identify the most relevant variables for the classification of the pyrolysis type of Biomass. Also, to evaluate if the pyrolysis type classes are sufficient to characterize this chemical process. The Random Forest algorithm was applied to identify which variables are relevant in the pyrolysis type classification process, obtaining an accuracy around 97%. It was identified that the most important variables are: Average residence time in the reactor for the gas and carrier, Percentage of ash-free dry basis carbon in the feedstock, Average particle size in the reactor and Percentage of ash-free dry basis hydrogen in the feedstock. With the most relevant variables, the following clustering methods were used: k-means, pam, clear, diana, fanny, hierarchical, sound, sota, model. To evaluate the clustering methods, internal validation measures with Dunn's index metrics and silhouette were used. The validation measures indicated the hierarchical clustering and k-means with better results for number of groups greater than the number of existing pyrolysis classes. Thus, the dataset should be divided into a larger number of pyrolysis type groups, as considering only the available classes is too limited to characterize the pyrolysis type, since the unsupervised classification algorithms indicate the number of clusters as greater than or equal to five. [ABSTRACT FROM AUTHOR]

Land use and land cover maps are basic tools for a detailed understanding of spatial organization patterns, an essential tool for agro-environmental management. This article aimed to compare and evaluate the accuracy of the digital classification of land use and land cover for the Cerrado of Maranhão, using OLI (Operational Terra Imager) images from Landsat-8 and MSI (Multispectral Instrument) from Sentinel2. The analyzed images are from 2021, which were pre-processed, segmented and classified using the Random Forest algorithm. The analyzes showed that both classifications were qualified as "very good", obtaining Kappa Index (0.706) and Global Accuracy (76%) for the OLI images and Kappa Index (0.775) and Global Accuracy (84%) for the images MSI. The results obtained can support the planning and execution of new agricultural mapping and monitoring in the Cerrado, positively influencing the decision-making of research groups. [ABSTRACT FROM AUTHOR]

The research presented in this article analyzed the performance of the random forest algorithm in predicting the future return of the main indices of the largest stock exchanges in the world, through historical trading prices. A sample composed of the daily quotes of 35 indices of the largest stock exchanges in the world from 2001 to 2019 was used. In addition to the random forest algorithm, models were estimated, based on the decision tree algorithm and using the logistic regression technique. The models were estimated considering maximum and closing prices, as well as the complete period and its division into sub-periods. The results indicated that the performances of the estimated models were superior to the market average, and the random forest presented the best results. All models trained on the maximum prices of the indices performed better than those trained on closing prices. In addition, the subperiod models performed better for the random forest. The efficiency of markets in the weak form has been questioned in the contemporary context of the rise of the use of artificial intelligence (AI) algorithms for forecasting in finance. The study is relevant as it contributes to the literature on the use of AI algorithms in forecasting asset prices in the financial market. The main indices of the largest stock exchanges in the world were analyzed, generating general subsidies that can help guide future research in the area. [ABSTRACT FROM AUTHOR]

Technological advances allow new content to be created and be available via Web every minute, providing great progress in several areas. However, this availability also brings drawbacks in the Educational field. It is noteworthy that the excess of materials/content makes teaching-learning process difficult due to the high time spent in searching for content that meets the needs of the users. In this sense, new methods to identify educational content, in videos, for example, need to be developed. From this perspective, it can be seen that significant differences are identified in the comments provided by users on educational videos, thus indicating the potential for using them in the process of selecting these types of videos. In this context, the present work analyzes and collects comments from 500 videos of the Youtube platform, being 250 educational and 250 non-educational, and uses Text Mining and Machine Learning techniques to develop a classification model that, based on the most frequent words of comments on videos, categorize them as educational or non-educational. Thus, we provide a mechanism that filters videos according to their class and returns to the user only videos with educational content. Results demonstrate that it is possible to classify educational and non-educational videos with an accuracy rate of 91.30%, when using the Random Forest classifier. Furthermore, due to the promising results, we developed SysVidEduc, an API that uses the comments from Youtube videos and automatically classifies them as educational or non-educational. [ABSTRACT FROM AUTHOR]

The health-disease process encompasses factors beyond genetic and biological susceptibility, but also includes variables linked to social and economic conditions that can lead to health vulnerability. The expanding situation of COVID-19 in Brazil has demonstrated how social inequalities affect this health-disease process; thus, evaluating such disparities can help the country confront the disease. The objective of this article was to establish an index to assess the situation of social vulnerability to COVID-19. From the 12 selected variables, the modeling identified those the predicted the occurrence of COVID-19 in the State of Goiás and the Federal District. For this, two machine learning algorithms were tested: Random Forest and XGBoost. The results indicated the most predictive variables were income status, the total hospitalizations for ailments classified as very vulnerable, and the percentage of the population working informally. Therefore, approximately 23% of the municipalities were classified with high to very high vulnerability. [ABSTRACT FROM AUTHOR]