شبکه‌های عصبی مصنوعی در پیش‌بینی ارزش طول عمر مشتریان تجارت الکترونیک: نقش شاخص‌های کلیدی عملکردی رضایتمندی مشتریان

Purpose: The rapid advancement of machine learning has enabled innovative applications across various business domains. This research explores the potential of neural networks to enhance customer relationship management by accurately predicting Customer Lifetime Value (CLV). By leveraging key performance indicators, including Net Promoter Score (NPS) and Customer Effort Score (CES), as input features, a neural network model is developed. This model aims to establish a predictive relationship between these metrics and CLV, providing valuable insights for businesses to optimize customer retention and revenue generation strategies. The proposed approach offers a scalable and customizable solution, enabling organizations to tailor the model to their specific needs and leverage the power of AI to drive data-driven decision-making.

Methodology: The data was collected in 2023 from customers of an Iranian e-commerce platform while ensuring the security of customer privacy and business-competitive confidentiality. The dataset includes 8,000 customer profiles with variables such as Customer Lifetime Value (CLV), Promoter Score (PS), Customer Effort Score (CES), and unique 16-digit customer identifiers. The model is a sequential neural network with dense and regularized layers, optimized through hyperparameter tuning. The model’s performance is evaluated on a 10% test set and a 10% validation set, using metrics such as Mean Absolute Error (MAE), R² (coefficient of determination), and other criteria. Additionally, the artificial neural network model’s performance is compared with two baseline models: linear regression and decision tree. Cross-validation is employed for model validation. Finally, the model is interpreted using SHAP and PFI methods, analyzing the importance of each variable in explaining Customer Lifetime Value.

Findings: The results indicate that the artificial neural network model, with 2 input neurons in the first layer, 5 hidden neurons in the second layer, 7 hidden neurons in the third layer, and 1 output neuron, effectively captures the complex and nonlinear relationships between Promoter Score (PS), Customer Effort Score (CES), and Customer Lifetime Value (CLV). The model achieves a coefficient of determination (R²) of 0.934 and a Mean Absolute Error (MAE) of 142.47, alongside several other performance metrics. The stable performance without overfitting, as validated through 10-fold cross-validation over 10 rounds, with an early stopping point around the fifth epoch, highlights the model's generalization capability. This allows for robust prediction of Customer Lifetime Value based on PS and CES. The model's ability to capture nonlinear relationships in the data outperformed baseline models (linear regression and decision tree), demonstrating the strength of the neural network. Furthermore, it was revealed that PS explains more variance in CLV than CES does.

Conclusion: This study demonstrated how effectively artificial neural networks can identify hidden patterns within key performance indicators. It highlights the capability of artificial neural networks to predict Customer Lifetime Value (CLV) in e-commerce, enabling the prediction of CLV using Promoter Score (PS) and Customer Effort Score (CES), along with facilitating precise customer segmentation, resource allocation, and strategic growth. Future research could improve CLV prediction by exploring additional datasets and extending the model, as well as investigating the potential of other machine learning algorithms.