Weighted logistic regression for large-scale imbalanced and rare events data

Maalouf, Maher; Siddiqi, Mohammad A. M.

doi:10.1016/j.knosys.2014.01.012

Cited by 65 publications

(56 citation statements)

References 30 publications

Supporting

Mentioning

Contrasting

Unclassified

Order By: Relevance

“…Tables and summarize the computational results for the 3 methods (TR‐IRLS, RE‐WLR, and TR‐PC) with the balanced and imbalanced training sets, including their optimal λ and accuracy. A comparison between the computational time efficiency (CPU time) between the algorithms is not included here because no statistical significance is found and the CPU times are comparable to the ones presented by Maalouf and Siddiqi …”

Section: Resultssupporting

confidence: 87%

“…Thus, to obtain consistent estimators, the likelihood is multiplied by the inverse of the fractions. Recently, Maalouf and Siddiqi developed the LR truncated Newton method for large‐scale imbalanced and REs data using the RE‐WLR algorithm and showed that the RE‐WLR algorithm outperforms the regular LR in terms of accuracy.…”

Section: Logistic Regression In Imbalanced and Res Datamentioning

confidence: 99%

“…Similar to TR‐IRLS termination criteria, the maximum number of iterations in Algorithm 1 is made 30, and for the deviance threshold, ε 1 , the value 0.01 is found to be adequate for good accuracy and speed. Algorithms 2 and 3 of the linear CG code for parameter and bias estimation, respectively, can be found in Maalouf and Siddiqi . In these algorithms (2 and 3), the CG iterations are set to 200 iterations while ε 2 and ε 3 are set to a value of 0.005.…”

Section: The Tr‐pc Algorithmmentioning

confidence: 99%

“…More recently, the prior correction method is used to predict natural hazards, especially landslide susceptibility . Maalouf and Siddiqi developed the LR truncated Newton method for large‐scale imbalanced and REs data using the rare events weighted logistic regression (RE‐WLR) algorithm.…”

Section: Introductionmentioning

confidence: 99%

See 3 more Smart Citations

Logistic regression in large rare events and imbalanced data: A performance comparison of prior correction and weighting methods

Maalouf

Homouz

Trafali̇s

2017

Computational Intelligence

Self Cite

View full text Add to dashboard Cite

The purpose of this study is to use the truncated Newton method in prior correction logistic regression (LR). A regularization term is added to prior correction LR to improve its performance, which results in the truncated‐regularized prior correction algorithm. The performance of this algorithm is compared with that of weighted LR and the regular LR methods for large imbalanced binary class data sets. The results, based on the KDD99 intrusion detection data set, and 6 other data sets at both the prior correction and the weighted LRs have the same computational efficiency when the truncated Newton method is used in both of them. A higher discriminative performance, however, resulted from weighting, which exceeded both the prior correction and the regular LR on nearly all the data sets. From this study, we conclude that weighting outperforms both the regular and prior correction LR models in most data sets and it is the method of choice when LR is used to evaluate imbalanced and rare event data.

show abstract

Section: Resultssupporting

confidence: 87%

Section: Logistic Regression In Imbalanced and Res Datamentioning

confidence: 99%

Section: The Tr‐pc Algorithmmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

See 2 more Smart Citations

Logistic regression in large rare events and imbalanced data: A performance comparison of prior correction and weighting methods

Maalouf

Homouz

Trafali̇s

2017

Computational Intelligence

Self Cite

View full text Add to dashboard Cite

show abstract

“…It has proven to minimize the generalization error of models. Auto adapting regularization parameters and applying a penalized logic regression based active learning to multi-class problems is suggested for future research (Maalouf & Siddiqi, 2014). Kernel methods transform data into higher dimensional space in contrast to linear classifiers that are implemented directly on data in its original space.…”

Section: Logistic Regressionmentioning

confidence: 99%

Text Classification Techniques: A Literature Review

Thangaraj

Sivakami

2018

IJIKM

View full text Add to dashboard Cite

Aim/Purpose: The aim of this paper is to analyze various text classification techniques employed in practice, their strengths and weaknesses, to provide an improved awareness regarding various knowledge extraction possibilities in the field of data mining. Background: Artificial Intelligence is reshaping text classification techniques to better acquire knowledge. However, in spite of the growth and spread of AI in all fields of research, its role with respect to text mining is not well understood yet. Methodology: For this study, various articles written between 2010 and 2017 on “text classification techniques in AI”, selected from leading journals of computer science, were analyzed. Each article was completely read. The research problems related to text classification techniques in the field of AI were identified and techniques were grouped according to the algorithms involved. These algorithms were divided based on the learning procedure used. Finally, the findings were plotted as a tree structure for visualizing the relationship between learning procedures and algorithms. Contribution: This paper identifies the strengths, limitations, and current research trends in text classification in an advanced field like AI. This knowledge is crucial for data scientists. They could utilize the findings of this study to devise customized data models. It also helps the industry to understand the operational efficiency of text mining techniques. It further contributes to reducing the cost of the projects and supports effective decision making. Findings: It has been found more important to study and understand the nature of data before proceeding into mining. The automation of text classification process is required, with the increasing amount of data and need for accuracy. Another interesting research opportunity lies in building intricate text data models with deep learning systems. It has the ability to execute complex Natural Language Processing (NLP) tasks with semantic requirements. Recommendations for Practitioners: Frame analysis, deception detection, narrative science where data expresses a story, healthcare applications to diagnose illnesses and conversation analysis are some of the recommendations suggested for practitioners. Recommendation for Researchers: Developing simpler algorithms in terms of coding and implementation, better approaches for knowledge distillation, multilingual text refining, domain knowledge integration, subjectivity detection, and contrastive viewpoint summarization are some of the areas that could be explored by researchers. Impact on Society: Text classification forms the base of data analytics and acts as the engine behind knowledge discovery. It supports state-of-the-art decision making, for example, predicting an event before it actually occurs, classifying a transaction as ‘Fraudulent’ etc. The results of this study could be used for developing applications dedicated to assisting decision making processes. These informed decisions will help to optimize resources and maximize benefits to the mankind. Future Research: In the future, better methods for parameter optimization will be identified by selecting better parameters that reflects effective knowledge discovery. The role of streaming data processing is still rarely explored when it comes to text classification.

show abstract