Scholarlink

The paper intends to investigate a comparison of different machine learning (ML)-based classification models used to categorize the consumer complaint narrative from the Consumer Financial Protection Bureau (CFPB) dataset. Two different ML algorithms were used for comparison: logistic regression and MNB. To attain the highest performance, one of the deep learning (DL) algorithms was enclosed in which the bidirectional LSTM model was compared with a pre-trained GLOVE embedding-based bidirectional LSTM model. In this work, it is observed that logistic regression shows significant improvement as compared to MNB. To improve the accuracy of the classification problem, a pre-trained GLOVE embedding-based bidirectional LSTM model is incorporated into the model, bidirectional LSTM model without pre-trained embedding provides the highest testing accuracy of 95.21%, as compared to with pre-trained embedding model. It is suggested that the model with pre-trained may not increase the performance. Thus, the bidirectional LSTM model without any pre-trained embedding is suitable for providing the best result.

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Prediction of Consumer Complaints Using Various Classification Models with Pre-trained Embedded Layer

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • Debarati Nath,
  • Joseph Roy,
  • Sukanya Ghosh,
  • Sayantani Bhattacharya

摘要

The paper intends to investigate a comparison of different machine learning (ML)-based classification models used to categorize the consumer complaint narrative from the Consumer Financial Protection Bureau (CFPB) dataset. Two different ML algorithms were used for comparison: logistic regression and MNB. To attain the highest performance, one of the deep learning (DL) algorithms was enclosed in which the bidirectional LSTM model was compared with a pre-trained GLOVE embedding-based bidirectional LSTM model. In this work, it is observed that logistic regression shows significant improvement as compared to MNB. To improve the accuracy of the classification problem, a pre-trained GLOVE embedding-based bidirectional LSTM model is incorporated into the model, bidirectional LSTM model without pre-trained embedding provides the highest testing accuracy of 95.21%, as compared to with pre-trained embedding model. It is suggested that the model with pre-trained may not increase the performance. Thus, the bidirectional LSTM model without any pre-trained embedding is suitable for providing the best result.