From Network Experience to Subscriber Retention: An Explainable AI Framework for Mobile Operators

1

From Network Experience to Subscriber Retention: An Explainable AI Framework for Mobile Operators

arXiv:2606.04838v1 [cs.NI] 3 Jun 2026

Faris B. Mismar, Senior Member, IEEE, Abdol Saleh, Ivan Maxmillian Putra Pasaribu, and Suhelmy Syaifuddin

Abstract—This article presents a framework for the prediction of subscriber churn in mobile operators also known as telecommunication operators (or telcos). This framework covers relevant aspects of data-driven approaches using explainable artificial intelligence and machine learning. To demonstrate the robustness of the framework, we implement it on real data from one of the globally leading telcos with tens of millions of subscribers and show results and actionable insights confirming the usefulness and longevity of the framework. Our results suggest that subscriber quality of experience (QoE) indicators provide stronger churn signals than traditional network counters alone, reinforcing the need for QoE-centric analytics in modern operations in telcos. We conclude with future research directions for improving churn predictability and operational deployment. Index Terms—Churn, framework, insights, artificial intelligence, machine learning, qoe, network, big data.

I. I NTRODUCTION The surge in smartphone subscriptions has been fueled by the widespread availability of technology, reduced device costs, application-centric digital ecosystems, and the growing reliance on mobile connectivity for all facets of life. Telecommunication operators (or “telcos”) are increasingly encountering flat or saturated mobile penetration levels (i.e., the percentage of a population that has active subscriptions). In this environment, sustaining customer loyalty becomes a strategic priority and a core competitive capability for telcos. Telcos continue to simultaneously operate several generations of wireless communications using the second (2G), third (3G), and fourth (4G) generation of wireless communications. Deployments of the fifth generation of wireless communications (5G) continue due to demands for much higher data rates and lower latency than what 4G offered. A drive to modernize networks and repurpose spectrum motivates telcos to phase out older technologies such as 2G and 3G [1]. Amidst this all, subscribers may choose to leave the telco, or churn. While churn in post-paid subscription is defined by events such as contract cancellation, we focus on the churn definition for pre-paid subscription, which is not as clear. In this context, churn is defined as a subscriber becoming non–revenue generating for a specified future duration, referred to as the threshold period (TP). Specifically, a prepaid subscriber is considered a churner if no account top-ups occur over consecutive months, resulting in zero revenue throughout the TP. A gross churner is any subscriber meeting this condition, while a net churner is a gross churner who does not resume any payment activity Faris B. Mismar (corresponding author) and Abdol Saleh are with Nokia Bell Labs Consulting; Ivan Maxmillian Putra Pasaribu and Suhelmy Syaifuddin are independent consultants.

before the TP expires. Churn can happen across all wireless communication technologies alike. Artificial intelligence (AI) and machine learning (ML) support data-driven decision processes by extracting patterns from large volumes of operational and customer data (i.e., big data). By learning from subscriber behavior, usage trends, and service interactions, churn prediction models can strengthen customer relationship management, inform smarter pricing and billing strategies, and optimize the allocation of constrained operational resources in advance—ultimately improving retention and safeguarding revenue streams. Explainable AI introduces a set of techniques that enable human users to understand, interpret, and trust the decisions produced by the ML model. It addresses the limitation of not knowing why a particular prediction was made by providing interpretable explanations, such as feature importance. In this article, we propose an explainable AI-driven framework designed to forecast net churn in advance in realistic telcos, enabling proactive and targeted intervention strategies and deriving insights about the churn at large. II. C HALLENGES Telcos striving to construct a churn model using an explainable AI framework may face several challenges. These challenges can be attributed to data, modeling pipeline (with machine learning operations (MLOps) being an example), and business adoption. In this section we talk about these challenges in adequate detail. A. Data challenges 1) Data silos: Data in telcos spans several organizations and domains. Each one of these may have their own data, which often sits in isolated systems creating a “siloed” dataset. As a result it becomes challenging to build a unified subscriber life cycle view. 2) Churn definition: The definition of a churning subscriber depends on several aspects such as pre-paid vs post-paid, period since last activity, and value leakage (i.e., moving to a lower tariff) vs churning out (i.e., leaving the telco). 3) ‘Perfect’ explainability: Given the many reasons why a subscriber can churn, it can be nearly impossible to capture all the data features that perfectly explain churn; thus the churn label is likely to be noisy. This is true because churn is associated with a set of limited service touchpoints providing limited observability. This noisy label can make the prediction performance of the model poor—due to overlapping classes— rendering the model unusable.

2

4) Data aggregation: Some learning features are collected per subscribers; others can be collected per serving base station. These aggregation granularities are rigid and cannot be modified. Moreover, data features can have heterogeneous temporal properties (e.g., collected at different time intervals). As a result creating a key that can be used to join the features into forming a “design matrix” [2] can be challenging. 5) Regulatory challenges: Some telcos are bound by regulations. For example, data is prohibited from leaving the sovereign land where the telco is based. Challenges intensify when the data engineering and data science teams are located in different geographies. B. Modeling challenges 1) Balanced learning: Highly imbalanced datasets are common in churn prediction because the number of churners is typically much smaller than the number of retained subscribers. While many ML algorithms can accommodate class imbalance through biasing the positive class weight or resampling, extreme imbalance may still degrade model stability and calibration. 2) Data drift and null ratios: Another important practical problem comes from data drift and null ratios. Data drift is the change in the underlying probability distribution of a given learning feature over a certain time window (i.e., nonstationarity). It can be measured using statistical quantities such as the population stability index (PSI). Null ratio is the ratio of missing values (e.g., NaN) in one learning feature to the total vector length. In several cases, lack of representative data from all touchpoints results in large missing values and forces some features to be dropped. Drift can happen due to many reasons such as: price changes, plan changes, new technology rollout, new site build, and campaigns from the competition. Evolving subscriber dynamics can also be a factor: While some of them are well understood, others remain latent and surface indirectly as noise in the data. Known drivers include the onboarding of new subscribers with preferences that differ from historical norms, changes in behavior among existing subscribers, environmental and contextual factors, and social influences such as peer, family, or community effects. Additionally, drift can happen due to windowing. Windowing uses the TP to construct the training data. For instance, if the TP spans three months (e.g., January through March of the current year), the training features must be derived from an earlier observation window, such as July through December of the previous year. Because of the windowing, training data is constructed using a fixed observation window per subscriber, while the churn label is defined as a forwardlooking indicator within the TP. During operation, inference is performed on a recurring monthly basis using the same trained model, with each cycle relying on newly transformed monthly data distributions. Throughout this process, both the model prediction scores and indicators of data drift are continuously monitored. This introduces an age-of-information effect [3]: as the input data becomes increasingly distant from the data used during training, model performance may degrade. 3) Model explainability: Telco decision makers look for actionable insights to improve the business. Explainability of

a model to address this need is to translate feature importances into specific feature-driven insights that can be used for churn mitigation actions. In this case, the framework provides explainability in terms of human-language insights, customer life-time value, and feature elasticity. 4) Productionization and time horizon: Transforming a validated model into a deployable, enterprise-grade solution while establishing reliable implementation under comprehensive model life-cycle governance poses a significant challenge for telcos. This process also encompasses continuous training, systematic validation, and periodic retraining to maintain performance and relevance over time. We call this validity time the time horizon of a model that represents its longevity. C. Business challenges 1) Actionability gap: which means that a strategy is needed to act upon the insights generated by the churn prediction model. Often, the concept of uplift (i.e., how many subscribers have been prevented from churning due to an action) is mentioned in this context. An actionability gap therefore is when an insight cannot be acted upon or does not lead to uplift. 2) Organizational silos: Telcos are structured around distinct domains (e.g., network engineering, marketing, customer care, and finance). Each domain has its own performance measures, systems, and leadership. Therefore, the ownership of a churn prediction model and how teams from the various domains interact to keep the model updated and relevant can be perceived as an extracurricular activity, causing its ownership to become fragmented, ambiguous, and ultimately deprioritized in favor of other domain-specific objectives. Now that we covered the practical challenges that the construction of a churn model in a telco can face, we next talk about the advantages of using explainable AI in the prediction of churn. III. O PPORTUNITIES FOR T ELECOM O PERATORS This section discusses some of the opportunities enabled by constructing churn models using an explainable AI framework. These opportunities are improved efficiency, observability, and decision intelligence. A. Efficiency AI can improve the classification performance of churn prediction. It can also accelerate the analysis and outputs to fraction of the time. These capabilities allow operators to respond faster to an imminent change in a subscriber behavior. B. Observability A churn model using explainable AI improves the observability task making the model outputs easier to monitor, interpret, and act upon. Here is how: 1) Churn propensity: The model outputs the churn labels of either retain (“0”) or churn (“1”) and the probability of churn (often called “churn propensity”). This value is a real number that lies between 0 and 1.

3

2) Reasons and explanations: The framework provides explanations for the predicted churn risk. These explanations reveal the main contributing factors explaining churn at large. They may include drivers from across different telco domains such as pricing, usage behavior, network experience, or customer care interactions. The explanations can also be aggregated to produce insights across customer segments. C. Decision Intelligence Informed operational decisions are facilitated by the churn model ability to provide transparent explanations and proper insights about subscriber retention. Here is what this means: 1) Improved decision transparency: Model outputs and explanations make predictions easier to understand. This increases confidence in the system across teams. 2) Improved customer retention strategies: The identified churn drivers by the order of feature importance help telcos design targeted retention actions. These actions may include new offerings, service improvements, pricing adjustments, or proactive customer support. IV. P ROPOSED F RAMEWORK After outlining the challenges and opportunities faced by telcos in developing predictive churn models using an explainable AI framework, we now delve into the framework we propose. A. Problem statement A major global telco serving tens of millions of prepaid subscribers across 2G, 4G, and 5G radio access networks (from several vendors) requires to build a solution that can 1) use up to 30% of the subscriber data of six months of history for training, 2) proactively predict net churners three months ahead of time (this is the TP) with a recall of at least 0.7 and F1 of at least 0.75 for the remaining (and unseen) subscriber data (of 70%), 3) explain the drivers of subscriber churn in the network (including the telco domain they belong to) through insights, and 4) be usable (i.e., for inference) for a time horizon of at least three months. This problem statement focuses exclusively on net churners; subscribers who return and generate revenue within the TP of 180 days are excluded. We propose a framework that addresses this problem statement. B. Framework Elements 1) Platform: Due to the anticipated size of data and in case of sovereignty requirements, a platform as a service (PaaS) provider can be employed to support scalable data processing and compute requirements through an instance residing in the country. 2) Data sources: Given that telco datasets span several domains and are often siloed as we stated earlier, identifying them is an essential element of the framework: 1) Billing and charging data, including subscriber tenure, tariff plans, and average revenue per user for the current and prior months.

2) Marketing and customer care data, covering exit surveys, trouble tickets, and call center interactions. 3) Network performance and utilization data, capturing KPIs across 2G, 3G, 4G, and 5G RAN as well as the core network. This includes a composite quality of experience (QoE) index and the subscriber’s most frequently served cell site. The QoE index aggregates metrics such as video quality, initial buffering delay, and streaming stalling events. 4) Geolocation and minimization of drive testing (MDT) data, providing session-level information such as serving cell identifier, device vendor, received signal power, throughput, and latency. For analytical clarity, data sources 1 and 2 are grouped as commercial data, the QoE index is categorized as experience data, and the remaining radio KPIs, core KPIs, and MDT data are referred to as network data. This categorization enables collective assessment of their respective contributions to churn and association to the proper telco domain. Additionally, metadata to translate the names of features into English (e.g., a lookup table) is required. 3) Data preprocessing: Given that data can have different keys (e.g., at the subscriber level vs at the serving base station level) or heterogeneous temporal properties (e.g., daily vs hourly), a processing step is has to construct the design matrix which later serves as an input to the churn prediction model. (i) The commercial and experience datasets have the subscriber identifier as the primary key while the network data has serving site identifier as the primary key. Joining these two types to form the design matrix requires a new definition: the favorite site identifier. Every subscriber has a favorite site identifier corresponding to the site that where the subscriber attempts the most sessions. (ii) To address the problem of heterogeneous temporal properties, we propose an aggregation by the subscriber key per learning feature to generate the minimum, median, maximum, and a few other percentiles in between— for robustness. Certain subscribers with a limited history (e.g., less than 30 samples) are tagged as cold subscribers and are dropped due to absence of statistical significance of these aggregation statistics. This effectively causes the number of columns to grow but captures enough variance for the prediction performance to be above a certain threshold. To ensure that feature importance remains meaningful, the generated features continue to use the original feature names with a suffix denoting the aggregation method chosen. We call this step feature densification. 4) Machine learning: An ML algorithm that works quite well for the churn prediction task is the Extreme Gradient Boosting (XGBoost) classifier [4]. XGBoost is particularly well suited for tabular datasets, performs robustly in the presence of missing values—once handled as described earlier—and has consistently achieved top rankings in data mining competitions due to its strong predictive performance [5]. Moreover, XGBoost can create explainable AI models due to its ability to provide feature importance. To train

4

Fig. 1. Modular architecture for churn prediction using machine learning.

the model successfully, an effective churn prediction model benefits strongly from a balanced dataset, in which the churn and non-churn classes are represented in equal or near-equal proportions. 5) Denoising heuristic: We have established that churn data can be noisy, which is one of the major challenges preventing the performance of a predictive model from meeting the desired classification performance targets. Since explainable AI models can provide probabilities of class membership for each subscriber, the subscribers with low probability values— or if the probability cannot be computed—are excluded from the prediction with a churn propensity of NaN, effectively introducing a third category for our classifier (i.e., churned, retained, or uncertain). This can happen for example due to noisy inputs or limited subscriber history. This way, the accuracy of the binary classes is protected against low confidence predictions which are dropped. This element of the framework is novel and is an integral part to ensure fulfilling the performance targets. 6) Maintenance: Due to the change in subscriber dynamics, non-stationarity, null ratios, and windowing effect, the ML model performance is expected to degrade over time. In this context, changes in the null ratio and PSI variations are monitored periodically as part of the MLOps efforts. The null ratio is computed per feature, with deviations beyond a predefined threshold signaling drift due to changes in missing value patterns. PSI, in contrast, captures shifts in feature distributions by comparing the probability distributions of training and inference data; it quantifies drift through the weighted difference of these distributions and their logarithmic ratios. PSI values are computed per feature and aggregated across all features. Any change in the null ratio or the PSI between the training an inference datasets, combined with the model performance falling below a predefined threshold, indicate the need for model retraining as part of the model maintenance. C. Value segments and churn A subscriber’s financial value to telco is defined by their profit margin contribution, commonly referred to as customer life-time value (CLV). This measure is derived from the monthly revenue, usage and other transactional costs, and the total number of months the user remains active. Throughout a subscriber’s life cycle, the distribution of this CLV by

subscribers and across the user base may vary, appearing skewed to the left, right, or remaining centered depending on the overall health of the portfolio. To capture the shifting value of subscriber equity, telcos can analyze subscriber cohorts using multi-month sliding windows. As illustrated in Fig. 2, this approach tracks dynamics ranging from the entry of new subscribers to their specific movements across cohorts with transition probabilities resembling a Markov chain. The total margin contribution to the telco fluctuates based on these internal equity shifts; overall value rises as highervalue microsegments (those to the “right”) grow and diminishes as the density of subscribers shifts toward lower-value segments (to the “left”). Ultimately, the service life cycle concludes with service termination, or churn, which can occur within any cohorts regardless of the subscriber’s value tier. To navigate this complex environment, telcos can leverage AIbased predictive models discussed in this article. Based on the churn propensity values, these models can predict tendencies to upgrades, downgrades, and terminations, allowing the telco to calculate the expected impact on its equity (i.e., gross profit margin) and strategic objectives. The value segments are constructed using the primary churn predictors, as we show in Section V. D. Architecture Our proposed framework entails a reusable modular architecture that streamlines the various elements mentioned earlier. This architecture includes four different blocks as shown in Fig. 1. 1) Data loading and transformation: This block ingests raw data from the data sources mentioned earlier, typically through database queries written in structured query language (SQL). However, direct SQL-based retrieval can be inefficient when operating on very large datasets. To address this, the query outputs are stored in more efficient formats, such as columnar files (e.g., Parquet). Executing this block produces multiple files that are stored in a PaaS-based object storage bucket. The feature densification data transformation step is then applied across the relevant learning features. This prepares the features for the subsequent steps. 2) Data joining, compression, and storing: In this step, the commercial and experience datasets are loaded and left-joined using the primary key, namely the subscriber identifier. Next,

5

TABLE I T OP -10 GROUPED LEARNING FEATURES BY CHURN PREDICTIVE POWER

Group Commercial data Experience data Network data

Fig. 2. Microsegments and dynamics of loyalty and value in the telco subscriber base.

the network data is then incorporated, with the subscriber’s favorite site identifier acting as a secondary join key. The result is a large, consolidated dataset that is stored in columnar format within a PaaS object storage bucket. This dataset, referred to the “design matrix”, is no longer time-based. The supertable is the augmentation of this design matrix by introducing the churn label as a supervisory signal. The churn label is based on the net churner definition evaluated over the TP defined in Section I. Specifically, the label indicates whether a subscriber is expected to churn at any point during the TP. The integration of the churn label requires careful alignment, not only on subscriber identity but also on the temporal relationship between historical data and the churn outcome as discussed earlier. This dataset, aggregated over a window is then stored in a columnar format. Owed to this module, the resulting learning task can therefore be framed as an example as: Given subscribers’ behaviors from July to December of last year, which subscribers are likely to churn during the subsequent January to March period? 3) AI model training: The supertable data is first cleansed ensuring no duplicate subscriber identifiers or subscribers with many missing value data points while preserving the class balance. An optional step of clustering (e.g., through K-Means clustering) or dimensionality reduction (e.g., through principal component analysis or factor analysis) can be implemented on the training data. The training of the XGBoost model takes place next. For the model training, a substantial portion of the supertable is divided into training and cross-validation sets, while the remaining data is reserved for testing and model scoring. The performance measures are the recall and the F1 score. For model calibration, training is performed using stratified K-fold cross-validation combined with randomized hyperparameter search over a broad parameter space. The optimization objective is to maximize a composite measure of both the area under the receiver precision-recall curve plus the F1 score. This allows us to find the decision threshold below

Ratio 90% 10% 0%

which we have a class (“0”) and above which it is a class (“1”). Stratification ensures that class balance is preserved across folds, enabling reliable hyperparameter selection and improved model performance. This model performance is known as the baseline performance. Finally, Shapley additive explanations (SHAP) [6] can be used to identify the impact the various features have on churn and retention in what is known as a “beehive plot”. 4) Inferencing and insights: The portion of the supertable remaining after the training and cross-validation splits are removed is used for evaluating the predictive capability of the model and for detecting overfitting or underfitting. Although the prediction task is described in forward-looking terms, all data involved in this evaluation originates from historical records. In effect, the model is assessed on its ability to “predict the past,” since true churn outcomes cannot be observed for future periods at the time of evaluation. Once the model demonstrates strong performance in predicting historical churn, the logical next step is to apply it to future churn prediction (i.e., in inferencing mode), which is on the true holdout (or unseen) dataset, representing the remaining 70% of the subscribers. This is where the true value of a churn prediction model emerges: providing telcos with sufficient advance notice to proactively engage subscribers who are at high risk of churning. Insights generation and explainability are done through sorting the top features by XGBoost feature importance, running SHAP, and using SHAP dependence plots for a principled approximation of feature elasticities (i.e., instead of the partial derivative definition which may be hard to compute). This, combined with the metadata, allows the generation of a SHAP-based insight in the syntax: “Subscribers with higher|lower metadata tend to have higher|lower churn risk (and the bins B have relative elasticities E).” We construct an MLOps pipeline automating the sequence of steps outlined in this architecture. This pipeline moves the ML workflow from raw data to a deployed model and its ongoing monitoring and report on its performance next. V. R ESULTS To test the relevance and validity of this framework, we implement a pipeline automating the architecture on a commercial PaaS using over 3,000 lines of source code. The data, as outlined in the problem statement, has approximately 800 learning features and require a storage bucket exceeding 250 TBs. The data cannot leave the country in which it resides. A. Top predictors Although model training incorporates commercial, experience, and network datasets, their contributions to churn

6

TABLE II M ODEL PERFORMANCE LONGEVITY

F1 Recall

Training period 0.8222 0.9118

Inference period 1 0.7591 0.7342

Inference period 2 0.7542 0.7279

Inference period 3 0.7487 0.7102

Fig. 3. Churn rate increases as value contribution declines.

prediction are not equal. We assess the relative importance of the various learning features, rank their predictive strength, map them to their data categories, and examine the underlying reasons for the observed trends as shown in Table I. Features derived from commercial datasets exhibit the strongest predictive power. Examples of these features are the subscriber tenure and recent account payment subscriber behavior (i.e., the telco revenue). Experience-related features rank second in importance. Measures such as perceived voice quality, video quality, and buffering delays (e.g., in games and streaming) can strongly influence churn, especially when service degradation occurs repeatedly at locations or times that are most critical to the subscriber. A notable observation from the analysis is that RAN busyhour performance counters–which are by design aggregated at the cell site level—exhibit limited predictive power for subscriber churn. This finding does not imply that network quality is irrelevant to churn prediction; rather, it highlights a measurement mismatch between network-centric KPIs and subscriber-perceived experience: Aggregated counters fail to capture spatial (i.e., cell center vs edge performance), temporal (e.g., peak vs off-peak), and device-level capability or variability experienced by individual subscribers. Moreover, MDT is sparse in nature due to subscribers behavior and mobility patterns. This sparsity can lead to one of two scenarios: 1) the null ratio in these learning features is too high or 2) the variance as explained by them would be too small. The former would cause these learning features to be dropped while the latter would cause them not to show among the top features by importance. As a result, we expect that models trained

primarily on RAN features perform only marginally better than random guessing when predicting churn. This allows us to state our observation that commercial and experience related data provide a closer proxy to perceived service quality. The heuristic algorithm removed approximately 3% of the “noisy” subscribers causing the recall to jump above the threshold defined in the problem statement. B. Windowing effect In Table II, we show the recall and F1 scores using the true churn data, which for inference purposes is obtained after the predictions are made. To study the model longevity, we observe the degradation in the performance measures over three inference periods, each spanning a TP of 90 days, defined using sliding windows that shift forward by one month. The degradation is expected due to the drift and age of information effect as discussed earlier. The model has a time horizon of 3 inference periods (or 5 months), after which the F1 score would fall below the cutoff. C. Value segments Value segments, as mentioned earlier, are constructed using the primary churn predictors (e.g., tenure, expenditure, usage, and experience). We depict the results in Fig. 3, where these segments are ranked ordinally from highest to lowest value. The top-ranked segments (e.g., Rank 1) exhibit high tenure and expenditure combined with optimal experience scores and relatively lower usage intensity; this profile correlates with high satisfaction, strong margin contribution, and high loyalty. Conversely, the bottom-ranked segments (e.g., Rank 36) display the opposite characteristics.

7

Fig. 3 illustrates the cumulative distribution of subscribers and churners. It demonstrates a clear inverse relationship between value segment ranks and their respective churn rates. As subscriber value declines, loyalty decreases and churn rates rise. We observe a distinct shift in the churn profile starting at the 20th segment, where the cumulative churn rate begins to accelerate. This trend steepens sharply beyond the 30th segment, indicating that the lowest-value cohorts are disproportionately driving the cumulative churn volume. D. Insights Based on the top predicting features from the commercial dataset and the SHAP-based insights syntax outlined in our proposed framework, here are three generated insights: • Subscribers with lower maximum subscriber tenure (in days) tend to have higher churn risk (31, −1.1%), (752.5, −34.2%). • Subscribers with lower median revenue in the current month (in USDs) tend to have higher churn risk (2, −25.8%), (7.19, −92.7%). • Subscribers with lower minimum quality of experience composite score and low median revenue (in USDs) tend to have higher churn risk ((20, 2), −12.4%), ((50, 7.9), −41.3%). These insights indicate that to mitigate churn, optimizing for QoE and retention and upsell campaigns should be considered. VI. F UTURE D IRECTIONS FOR AI-D RIVEN S UBSCRIBER R ETENTION Examples of future directions are prediction of value leakage and per-subscriber causal analysis. A. Value leakage prediction Value leakage is the gradual erosion of revenue as opposed to the hard and binary nature customer departure. Value leakage happens while the subscriber remains active. In telcos, this may manifest as a decline of revenue, reduced usage intensity, plan downgrades, increased discount dependency, or migration to lower-value bundles. Unlike churn, which is a discrete event, value leakage is continuous and often subtle, making it harder to detect but economically significant at scale.

not uniformly distributed across value segments that design and implementation of churn mitigation strategies must consider to drive optimal return on investment. We further showed how noisy churn data can be handled without a major loss in predictive power through a denoising heuristic. R EFERENCES [1] H. Remmert, “2G, 3G, 4G LTE Network Shutdown Updates,” Dec. 2025, Accessed on 2026-03-03. [Online]. Available: https: //www.digi.com/blog/post/2g-3g-4g-lte-network-shutdown-updates [2] I. J. Goodfellow, Y. Bengio, and A. Courville, Deep Learning. Cambridge, MA, USA: MIT Press, 2016. [Online]. Available: http: //www.deeplearningbook.org [3] S. Kaul, M. Gruteser, V. Rai, and J. Kenney, “Minimizing age of information in vehicular networks,” in 2011 8th Annual IEEE Communications Society Conference on Sensor, Mesh and Ad Hoc Communications and Networks, 2011, pp. 350–358. [4] T. Chen and C. Guestrin, “XGBoost: A Scalable Tree Boosting System,” Mar. 2016. [5] F. B. Mismar, A. Alammouri, A. Alkhateeb, J. G. Andrews, and B. L. Evans, “Deep learning predictive band switching in wireless networks,” IEEE Transactions on Wireless Communications, vol. 20, no. 1, pp. 96– 109, 2021. [6] L. S. Shapley, “Notes on the n-Person Game – II: The Value of an nPerson Game,” Santa Monica, CA, RAND Corporation., Aug. 1951. [7] F. B. Mismar and J. Hoydis, “Unsupervised learning in next-generation networks: Real-time performance self-diagnosis,” IEEE Communications Letters, vol. 25, no. 10, pp. 3330–3334, 2021. Faris B. Mismar is a Senior Principal Consultant and a Distinguished member of Technical Staff at Nokia Bell Labs. He is also an Adjunct Associate Professor of Electrical and Computer Engineering at the University of Texas at Dallas. An accomplished technical leader, his research has directly influenced 3GPP conditional handovers procedures (3GPP TS 28.313) and facilitated the creation of production O-RAN rApps and xApps. In his 25-year career, he has led or contributed to over 100 large-scale client-facing projects worldwide. He is a Senior Member of the IEEE and has served as an Associate Editor of IEEE T RANSACTIONS ON M ACHINE L EARNING IN C OMMUNICATIONS AND N ETWORKING since 2025. His research interests include machine learning, network optimization, and wireless communications. Abdol Saleh is a Principal Consultant and a Distinguished Member of Technical Staff at Nokia Bell Labs. With over three decades of experience in the telecommunications industry, he specializes in digitalization technologies, AI/ML and big data platforms. His expertise includes network automation, industrial and telecommunication econometrics analysis, customer experience and value management. Abdol holds a PhD in Industrial Engineering and Operations Research from Lehigh University (1988). He has continued to expand his expertise with advanced certifications, including a Professional Certificate in Data Engineering from MIT (2024), an Applied Data Science Certificate from MIT (2022), and a Smart Industry Readiness Index CSA Certification (INCIT, 2021), reinforcing his commitment to driving digital transformation in telecommunications and beyond.

B. Causal analysis The microsegmentation insights and analysis can be advanced beyond macro-level insights by estimating individualized cause–effect relationships (e.g., “customers with high complaints churn more”). Instead of identifying correlations across large segments, causal analysis seeks to understand the effects at the per-subscriber level. Causal inference [7] and counterfactual modeling techniques are used in this space. VII. C ONCLUSIONS This paper demonstrated that effective prediction of subscriber churn in a telco scenario requires an AI-enabled analytics framework. This framework empowers telcos with the ability to predict the subscriber churn propensity, derive actionable insights, and align with the customer life-time value over their life cycle. Specifically, we showed that churners are

Ivan Maxmillian Putra Pasaribu is a Senior Officer of Data Science in the Data Analytics and AI division at a telecommunications operator in Indonesia. His work focuses on the application of machine learning and artificial intelligence to network analytics, customer experience management, and business process optimization in telecommunications. He has led AI use case implementation and cross-functional collaboration to advance AI adoption across the company. Ivan holds a Bachelor of Engineering in Electrical and Electronics Engineering from the University of Indonesia (2023). His research interests include machine learning, data analytics, and network optimization. Suhelmy Syaifuddin is the Vice President and Head of Data Analytics and AI/ML at a leading telecommunications operator in Indonesia. With over two decades of experience spearheading transformative initiatives in the telecommunications and data sectors, his leadership focuses on the end-toend development of scalable agentic AI solutions and customer experience excellence. His professional background includes the development of diverse SaaS products, alongside the management of large-scale IT and analytic projects. His current research and professional interests are centered on scalable machine learning and agentic customer experience intelligence.