“…Sabo et al (2021) used clustering approaches in judicial decisions to obtain information about important factors in the air transport, such as service failures. Lei et al (2017) used supervised models: Naive Bayes, Support Vector Machines and Random Forest to classify documents of Chinese court decisions. The SVM was the most accurate among them.…”
Section: Importance Of Machine Learning Approaches In Different Areas...mentioning
In Brazil, one of the most harmful costs for airlines is the number of lawsuits filed against them. It is a problem that can affect its operations, reduce the entry of new competitors and create legal uncertainty in the country. This work seeks to highlight the factors which most contribute to the rise of judicial indemnities, discuss the most relevant issues and identify the best techniques to predict the indemnified values. The objective is to provide subsidies for airlines to mitigate the number of legal actions by using machine learning models. This research contributes by discussing one of the most relevant subjects in Brazilian air transport and comparing the machine learning models’ performance. The study is based on lawsuits between 2016 and 2021 using the companies’ data. The performance of Naive Bayes, Random Forest, Support Vector Machines, and Multinomial Logistic Regression models are evaluated through the accuracy, area under the ROC curve, and confusion matrix. The results showed better predictive power for Random Forest and Logistic Regression. The latter showed that flight delays, cancellations, and airline faults have a negative effect on indemnities. The above-average compensation is a tendency in some states, being the moral damage awarded to customers the main cause of higher compensation.
“…Sabo et al (2021) used clustering approaches in judicial decisions to obtain information about important factors in the air transport, such as service failures. Lei et al (2017) used supervised models: Naive Bayes, Support Vector Machines and Random Forest to classify documents of Chinese court decisions. The SVM was the most accurate among them.…”
Section: Importance Of Machine Learning Approaches In Different Areas...mentioning
In Brazil, one of the most harmful costs for airlines is the number of lawsuits filed against them. It is a problem that can affect its operations, reduce the entry of new competitors and create legal uncertainty in the country. This work seeks to highlight the factors which most contribute to the rise of judicial indemnities, discuss the most relevant issues and identify the best techniques to predict the indemnified values. The objective is to provide subsidies for airlines to mitigate the number of legal actions by using machine learning models. This research contributes by discussing one of the most relevant subjects in Brazilian air transport and comparing the machine learning models’ performance. The study is based on lawsuits between 2016 and 2021 using the companies’ data. The performance of Naive Bayes, Random Forest, Support Vector Machines, and Multinomial Logistic Regression models are evaluated through the accuracy, area under the ROC curve, and confusion matrix. The results showed better predictive power for Random Forest and Logistic Regression. The latter showed that flight delays, cancellations, and airline faults have a negative effect on indemnities. The above-average compensation is a tendency in some states, being the moral damage awarded to customers the main cause of higher compensation.
“…Em [Lei et al 2017] é feita uma classificação de julgamentos em 13 categorias relacionadas à qualidade de produtos, com base no padrão legal da divisão da indústria chinesa. Na etapa de pré-processamento foram removidas as stopwords não apenas da língua chinesa, mas também foram identificadas palavras do meio jurídico que estão sempre presentes e tem pouco significado, além disso foi realizada a tokenização em TF-IDF e utilizadas 3 técnicas de redução de features: mínima frequência de documentos, PCA e SVD.…”
Section: Trabalhos Relacionadosunclassified
“…O trabalho proposto também fará uma classificação binária assim como [Bahgat et al 2018] e [Aletras et al 2016], e utilizará a tokenização em TF-IDF como [Bahgat et al 2018] e [Lei et al 2017], visto que o TF-IDF também considera a frequência de documentos em que o token aparece ao invés de apenas contabilizar a quantidade de ocorrências total do mesmo como ocorre no BOW. Será utilizado o modelo híbrido de tokenização em 1-gram e 2-gram assim como no modelo de melhor resultado obtido por [Caccamisi et al 2020].…”
Section: Trabalhos Relacionadosunclassified
“…Será utilizado o modelo híbrido de tokenização em 1-gram e 2-gram assim como no modelo de melhor resultado obtido por [Caccamisi et al 2020]. Não será aplicado nenhum método de redução ou seleção de features como ocorreu em [Bahgat et al 2018] e [Lei et al 2017], visto que os algoritmos de classificação escolhidos (NB, SVM, LR e RF) possuem um bom desempenho para altas dimensões [Caruana et al 2008] [Palatucci and Mitchell 2007] [Ukey and Alvi 2012. A remoção de stopwords corriqueiras do meio jurídico feita em [Lei et al 2017] pode ser considerada o inverso do que está sendo proposto com as features customizadas, pois ao invés de detectar as palavras menos importantes para serem retiradas são detectadas as mais importantes para elevar seu peso no modelo, algo que não é feito por nenhum dos estudos apresentados e que é um diferencial deste trabalho.…”
Durante um processo de acreditação de organismos de inspeção, relatórios precisam ser revisados por uma equipe técnica do Inmetro. Utilizando as respostas dadas nesses relatórios, este estudo propõe um procedimento para análise automática dos relatórios de acreditação do Inmetro. O objetivo é categorizar os componentes dos relatórios em adequados ou inadequados, evitando a necessidade de revisão manual. Além disso, pretende-se aumentar a eficiência dos modelos utilizando features customizadas, que seriam elementos identificados como importantes para que uma resposta seja considerada adequada. Nos experimentos, o SVM foi o algoritmo com melhor resultado para o problema e a utilização de features customizadas melhorou o desempenho final a depender da pergunta e do algoritmo utilizado.
“…ZH Lin researched of criminal case semantic feature extraction method based on the Convolutional Neural Network [2]. M Lei based on Machine Learning algorithms to implement automatic classification of Chinese Judgment Documents [3]. YL Chen designed a textmining-based method that allows the general public to use everyday vocabulary to search for and retrieve criminal judgments [4].Similar researches mostly make statistics on the contents of referees' documents through statistical knowledge and find out some superficial rules in the documents.…”
In recent years, along with the improvement of population quality and the advancement of the rule of law society, the market for legal services in the middle and low-end markets has continued to expand, and legal advice has become widespread in daily life. In the process of legal services, the legal provisions play an important role in the lawyer's decisionmaking. Meanwhile, the historical cases can help the lawyers and the parties to draw lessons from similar cases. However, with the increasing number of judicial documents, it is becoming increasingly difficult to summarize and learn from history. Therefore, this paper proposes a sentencing interval prediction model of criminal cases based on convolutional neural network, and through the method of multi-core convolution, greatly enhances the generalization ability and prediction performance of the model. The experimental analysis of real criminal case verdict verifies that the model is more effective than other classification prediction algorithms.
scite is a Brooklyn-based organization that helps researchers better discover and understand research articles through Smart Citations–citations that display the context of the citation and describe whether the article provides supporting or contrasting evidence. scite is used by students and researchers from around the world and is funded in part by the National Science Foundation and the National Institute on Drug Abuse of the National Institutes of Health.