An Insurance Sales Prediction Model Based on Deep Learning

Haiping WangRevue 2020 · DOI

Abstract: The analysis and prediction of insurance sales enable insurance companies to construct effective and efficient marketing models, conduct targeted promotion campaigns, and expand the range of target consumers, thereby increasing the direct and indirect profits. This paper improves the long short-term memory (LSTM) network into an effective prediction model for insurance sales. Firstly, an original database was created on the historical data of insurance sales, followed by the identification of the linear and nonlinear factors affecting insurance sales. Next, the redundant data were removed from data sequences through nondimensionalization and gray correlation calculation. Finally, multiple linear regression (MLR) was carried out based on LINEST function, completing the prediction of insurance sales. Simulation results show that the improved LSTM network outperformed the other prediction models, indicating that the prediction effect can be improved by data processing through multiple residual predictions.

Introduction

Predicting insurance sales is a critical yet challenging task for insurance companies aiming to optimize marketing strategies and maximize profits. The difficulty stems from the complex interplay of linear market trends and highly nonlinear factors, such as seasonal cycles, holiday distributions, and flash promotions. Traditional models often fail to pinpoint demand spikes accurately, leading to inefficient resource allocation. Furthermore, standard Artificial Neural Networks (ANNs) frequently encounter issues like vanishing or exploding gradients when processing the high-dimensional datasets typical of the insurance industry. This paper proposes an improved Long Short-Term Memory (LSTM) network that integrates statistical preprocessing to overcome these hurdles.

Methodology

The proposed framework utilizes a multi-stage pipeline that combines classical statistical methods with deep learning to enhance prediction accuracy.

Proposed multi-stage pipeline for insurance sales prediction.

The process begins with Gray Correlation Analysis to identify and rank factors (e.g., holidays, promotions) that significantly impact sales, allowing for the removal of redundant data. Subsequently, Multiple Linear Regression (MLR) via the LINEST function is employed to model linear relationships, preventing the deep learning model from falling into local minimum traps. The core innovation lies in the Improved LSTM with Multiple Residual Prediction. Instead of a single-pass prediction, the model iteratively predicts residual sequences—the errors between reference and comparison data—and incorporates these optimized residuals back into the final output to fine-tune the results.

Data Analysis and Insights

The study analyzed four years of billing data (2016–2019) from a major insurance provider, covering 207 types of insurance across 18 classes. Data from 2016–2018 served as the training set, while 2019 data was used for testing.

Historical insurance sales trends showing seasonal fluctuations across property and personal insurance types.

A key finding from the data analysis is the distinct seasonality of insurance sales. As shown in the figure above, sales follow a roughly sinusoidal distribution, peaking in the first (Q1) and fourth (Q4) quarters, with notable lulls during the middle of the year. By identifying these patterns and using Gray Correlation to streamline features, the model achieves faster convergence and higher accuracy compared to models using raw, unprocessed data.

Results and Discussion

The hybrid model was benchmarked against a traditional LSTM and a Backpropagation Neural Network optimized by a Genetic Algorithm (GA-BPNN). The results demonstrate significant improvements:

  • Accuracy Boost: The error rate dropped from 26.77% (GA-BPNN) to just 8.25% with the proposed model.
  • Stability: The improved model reduced the error rate by 12.5% compared to a standard LSTM, showing superior performance in handling large fluctuations in sales sequences.
  • Consistency: The mean error remained stable at approximately 16 items per week, even amidst changes in promotion formats and holiday schedules.

Conclusion

By combining the strengths of Gray Correlation Analysis, MLR, and an iterative residual-optimizing LSTM, this research provides a robust framework for insurance sales forecasting. While the model excels at capturing seasonal trends and linear growth, future work will focus on improving its resilience to "black swan" events—sudden, unpredictable market shifts that fall outside historical patterns. Integrating real-time economic indicators could further enhance the model's predictive power in volatile environments.

Summary updated

Abstract

The analysis and prediction of insurance sales enable insurance companies to construct effective and efficient marketing models, conduct targeted promotion campaigns, and expand the range of target consumers, thereby increasing the direct and indirect profits. This paper improves the long short-term memory (LSTM) network into an effective prediction model for insurance sales. Firstly, an original database was created on the historical data of insurance sales, followed by the identification of the linear and nonlinear factors affecting insurance sales. Next, the redundant data were removed from data sequences through nondimensionalization and gray correlation calculation. Finally, multiple linear regression (MLR) was carried out based on LINEST function, completing the prediction of insurance sales. Simulation results show that the improved LSTM network outperformed the other prediction models, indicating that the prediction effect can be improved by data processing through multiple residual predictions.
BibTeX
@inproceedings{wang2020insurance,
  author = {Haiping Wang},
  title = {{An Insurance Sales Prediction Model Based on Deep Learning}},
  booktitle = {Revue},
  year = {2020},
  doi = {10.18280/ria.340309},
}
Figure