Scholarlink

<p>The key to personalized retail is to accurately predict consumer behavior, but traditional models can be problematic due to the large dimensionality of demographic data and non-linear relationships between demographics and behavior. In this paper, a new deep learning model is suggested, which applies a Convolutional Neural Network (CNN) to estimate the level of individual income and provide specific product suggestions. In contrast to the traditional tabular learners, our method converts normalized customer features to grayscale image matrices of size 20 × 10, allowing the CNN to learn the complex spatial features and latent behavioral patterns in the hybrid pooling layers. The algorithm is implemented in two combined steps: high-granularity income tier categorization and a recommendation engine that is powered by a purchase probability matrix. The experimental findings using a dataset of 980 people prove that the proposed model is much better than state-of-the-art benchmarks and has statistically significant accuracy of 93.06 in income prediction and 95 in recommendation success. These results highlight how the use of spatial feature extraction can be more effective in consumer analytics and offer a scalable pipeline to e-commerce real-time personalization.</p>

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

Predicting customer buying habits using convolutional neural network

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年5月20日

查看全文
  • Zhuang Lou,
  • Shuai Wang,
  • Xiaoyue Yu,
  • Wei Song

摘要

The key to personalized retail is to accurately predict consumer behavior, but traditional models can be problematic due to the large dimensionality of demographic data and non-linear relationships between demographics and behavior. In this paper, a new deep learning model is suggested, which applies a Convolutional Neural Network (CNN) to estimate the level of individual income and provide specific product suggestions. In contrast to the traditional tabular learners, our method converts normalized customer features to grayscale image matrices of size 20 × 10, allowing the CNN to learn the complex spatial features and latent behavioral patterns in the hybrid pooling layers. The algorithm is implemented in two combined steps: high-granularity income tier categorization and a recommendation engine that is powered by a purchase probability matrix. The experimental findings using a dataset of 980 people prove that the proposed model is much better than state-of-the-art benchmarks and has statistically significant accuracy of 93.06 in income prediction and 95 in recommendation success. These results highlight how the use of spatial feature extraction can be more effective in consumer analytics and offer a scalable pipeline to e-commerce real-time personalization.