← 研踪 PaperTrailarXiv:2606.04838 · v3
From Network Experience to Subscriber Retention: An Explainable AI Framework for Mobile Operators
首次提交 2026-06-03 21:08
版本更新 2026-11-04 22:10
雷达观测 2026-11-05 12:13
摘要
This paper presents an explainable AI-driven framework for predicting net churn among prepaid mobile subscribers. The framework formulates churn prediction as a machine learning task mapped onto an industry reference architecture. Moreover, it introduces a quantile-based feature densification transform that reconciles learning features of unequal length and granularity into a homogeneous design matrix, evaluates a label-aware truncation diagnostic that sets aside low-confidence subscribers, and generates automated English-language insights describing the impact of each feature on churn propensity. We implement and validate the framework on real production data from a globally leading telco with tens of millions of prepaid subscribers spanning approximately $800$ learning features and over $250$ TB of commercial, experience, and network data. As a diagnostic, setting aside roughly $3$% of the subscribers, selected using the true labels, lifts $F_1$ on the retained subscribers to the target. This is not an achievable operating performance on all subscribers. In our data and feature set, aggregated radio access network counters carried less predictive signal than commercial and experience indicators, despite their greater representation. The quality of experience index, which is derived from network measurements, ranked second in importance. Our results also suggest that churn risk rises as subscriber value declines, with high-value segments disproportionately exposed to gradual value leakage rather than outright attrition.
作者备注：14 pages, 13 figures, submitted to IEEE Access
观测到的版本
相邻记录可能跳过中间版本；首次发现 v2 及以上也会标为版本更新。
v3 · 2026-11-04 22:10 · 最新观测版本
From Network Experience to Subscriber Retention: An Explainable AI Framework for Mobile Operators
This paper presents an explainable AI-driven framework for predicting net churn among prepaid mobile subscribers. The framework formulates churn prediction as a machine learning task mapped onto an industry reference architecture. Moreover, it introduces a quantile-based feature densification transform that reconciles learning features of unequal length and granularity into a homogeneous design matrix, evaluates a label-aware truncation diagnostic that sets aside low-confidence subscribers, and generates automated English-language insights describing the impact of each feature on churn propensity. We implement and validate the framework on real production data from a globally leading telco with tens of millions of prepaid subscribers spanning approximately $800$ learning features and over $250$ TB of commercial, experience, and network data. As a diagnostic, setting aside roughly $3$% of the subscribers, selected using the true labels, lifts $F_1$ on the retained subscribers to the target. This is not an achievable operating performance on all subscribers. In our data and feature set, aggregated radio access network counters carried less predictive signal than commercial and experience indicators, despite their greater representation. The quality of experience index, which is derived from network measurements, ranked second in importance. Our results also suggest that churn risk rises as subscriber value declines, with high-value segments disproportionately exposed to gradual value leakage rather than outright attrition.
查看此版本解读 →查看此版本原文 ↗