Quantum geometric-entropic optimization for customer lifetime value prediction: convergence theory and an empirical study on transactional retail data

This study aims to segment customers using the application of the K-means clustering algorithm and predict customer churn using the random forest method. Transactional data were used, including the order date, customer name, region, logistics company, quantity bought, payment amount, and frequency bought. K-means clustering was applied to group customers into segments, while a random forest model was constructed to predict customer churn. K-means clustering could determine four customer segments with different purchasing habits. Random forest model could predict customer churn and could find that attributes such as payment value and region were the most significant to use while determining the probability of churn. Results of this study verify that employing K-means clustering and random forest simultaneously for customer segmentation and customer churn prediction is efficient and assists in obtaining considerable insights for precision marketing.

PurposeAs internet banking service marketing platforms continue to advance, customers exhibit distinct behaviors. Given the extensive array of options and minimal barriers to switching to competitors, the concept of customer churn behavior has emerged as a subject of considerable debate. This study aims to delineate the scope of feature optimization methods for elucidating customer churn behavior within the context of internet banking service marketing. To achieve this goal, the author aims to predict the attrition and migration of customers who use internet banking services using tree-based classifiers.Design/methodology/approachThe author used various feature optimization methods in tree-based classifiers to predict customer churn behavior using transaction data from customers who use internet banking services. First, the authors conducted feature reduction to eliminate ineffective features and project the data set onto a lower-dimensional space. Next, the author used Recursive Feature Elimination with Cross-Validation (RFECV) to extract the most practical features. Then, the author applied feature importance to assign a score to each input feature. Following this, the author selected C5.0 Decision Tree, Random Forest, XGBoost, AdaBoost, CatBoost and LightGBM as the six tree-based classifier structures.FindingsThis study acclaimed that transaction data is a reliable resource for elucidating customer churn behavior within the context of internet banking service marketing. Experimental findings highlight the operational benefits and enhanced customer retention afforded by implementing feature optimization and leveraging a variety of tree-based classifiers. The results indicate the significance of feature reduction, feature selection and feature importance as the three feature optimization methods in comprehending customer churn prediction. This study demonstrated that feature optimization can improve this prediction by increasing the accuracy and precision of tree-based classifiers and decreasing their error rates.Originality/valueThis research aims to enhance the understanding of customer behavior on internet banking service platforms by predicting churn intentions. This study demonstrates how feature optimization methods influence customer churn prediction performance. This approach included feature reduction, feature selection and assessing feature importance to optimize transaction data analysis. Additionally, the author performed feature optimization within tree-based classifiers to improve performance. The novelty of this approach lies in combining feature optimization methods with tree-based classifiers to effectively capture and articulate customer churn experience in internet banking service marketing.

Empirical analysis of tree-based classification models for customer churn prediction

The evolution of financial services in the digital era has enabled access to alternative data streams beyond traditional credit bureau records, opening new possibilities for credit scoring and loan default prediction. In both formal banking systems and emerging fintech platforms, the integration of behavioral and transactional financial data offers richer, more dynamic insights into borrower risk profiles. This shift has paved the way for machine learning (ML) models to enhance the accuracy, fairness, and scalability of credit assessment processes. This paper investigates the application of machine learning algorithms in credit scoring and loan default prediction, using behavioral signals such as spending patterns, payment timing, mobile usage and transactional data from bank accounts, e-wallets, and point-of-sale interactions. Supervised learning techniques like logistic regression, random forests, gradient boosting, and neural networks are benchmarked against traditional credit scoring models to assess predictive performance and generalization. Additionally, the paper examines the role of unsupervised clustering for segmenting borrower profiles and semi-supervised learning for scenarios with limited labeled data. Feature engineering methods, including temporal trend extraction, merchant categorization, and transaction frequency analysis, are discussed in detail. The paper also addresses challenges related to data privacy, class imbalance, and model interpretability highlighting techniques such as SHAP values and local interpretable model-agnostic explanations (LIME) to improve transparency in ML-driven decisions. By incorporating diverse data sources and advanced analytics, ML-based credit scoring systems offer enhanced precision in predicting defaults, expanding financial inclusion while reducing systemic risk. Case studies from microfinance, mobile lending, and digital banking underscore the real-world applicability of these models in low-data and high-risk environments.

Machine learning (ML) encompasses a diverse array of both supervised and unsupervised techniques that facilitate prediction, classification, and anomaly detection. Among the many fields of application for such techniques, customer churn prediction is a prominent one. In order to forecast customer switching, data scientists employ a variety of demographic, social, transactional, and behavioral variables and attributes. Unfortunately, many businesses in the United Kingdom still lack the comprehensive and adaptable consumer data required to perform accurate analyses. As a result, they often rely heavily on data produced by enterprise resource planning systems, which is primarily transactional in nature. Consequently, businesses are often limited to modeling and forecasting on transactional data alone and are unlikely to invest significantly in marketing research or other customer-related sources. Businesses are often limited to performing modeling and forecasting on transactional data that are most often not based on advanced techniques like recency, frequency and monetary (RFM) and ML. So, the major objective of the current work is to provide a mix of ML and RFM analysis techniques for churn prediction using mostly transactional data. The dataset was taken from the dataset search website containing online retail datasets. Every customer's RFM scores are computed based on the available data. A churn metric that indicates whether or not the customer has made a transaction in a limited time. Through this paper, different techniques are compared. We used K-means and DBSCAN clustering. By the end of this paper, it may be inferred that the act of dividing customers into six distinct clusters is a more practical and straightforward approach. Received: 29 June 2023 | Revised: 21 August 2023 | Accepted: 6 September 2023 Conflicts of Interest The author declares that she has no conflicts of interest to this work. Data Availability Statement The data that support the findings of this study are openly available in [Google Drive] at https://drive.google.com/file/d/1qme8WeYkmXWfWkLa87jG37owWPCsNY65/view?usp=drive_web

Sampling-based novel heterogeneous multi-layer stacking ensemble method for telecom customer churn prediction

Predicting customer churn in retail banking is essential for sustaining profitability. This study compares four supervised machine-learning models—Logistic Regression (LR), Support Vector Machine (SVM), Random Forest (RF), and Artificial Neural Network (ANN)—using the publicly available bank customer churn dataset from Kaggle (10,000 records, 18 attributes; publicly available at Kaggle repository. Data preprocessing included one-hot encoding for categorical variables, label encoding for gender, and feature selection via an ExtraTreesClassifier retaining nine informative predictors (e.g., age, credit score, balance). To address class imbalance (≈80% nonchurners vs. 20% churners), models were trained and evaluated with and without the Synthetic Minority Oversampling Technique (SMOTE), which was applied only to the training folds under stratified 5-fold cross-validation. Evaluation metrics comprised accuracy, precision, recall (for churn class), F1-score, ROC–AUC, and PR–AUC. RF achieved the best balance between recall (0.484 imbalanced; 0.619 balanced) and accuracy (0.867 imbalanced), while LR with SMOTE attained the highest recall (0.715) at the cost of reduced accuracy (0.718). Overall, the results highlight RF as the most robust model across both distributions and emphasize the importance of imbalance-aware evaluation in bank churn prediction.

Curse of dimensionality will occur if effective dimensionality reduction method is not applied in machine learning, especially in the telecom field. The existing researches on customer churn are still lack of a set of scientific, system theory and method and the single models methods for customer churn prediction also are unable to completely meet application needs. Therefore, it is important for theoretical and practical contribution to explore and study the customer churn prediction. Based on the substantive characteristics of customer churn in telecommunication, the indicator system of customer churn in telecommunication are studied in this paper. Firstly, we proposed a feature selection method based on pruning technique, which is called feature selection method based on orientation ordering pruning Method (OOPM). According to this algorithm attribute selection problem can be replaced by the pruning question of classifier combination and we structure an indicator system of customer churn. Secondly, in order to explore high-order statistical information in the properties, a feature extraction method based on Random Forest and Transduction (FE_RF&T) is proposed to extract multiple features from customer data. Experiments on the real data of telecom enterprise show that the feature selection method based on OOPM has more advantages than the feature selection method based on Random Forest, and compared to the PCA method the FE_RF&T method improves the performance of learning machine effectively.

This research explores e-commerce dynamics, focusing on the challenge of predicting customer churn using deep learning [65]. It integrates and analyses both textual and transactional data, including social media posts and customer feedback [59]. The approach uses an advanced deep learning model, involving data collection, pre-processing, and feature extraction [40]. Novel methods fuse data to create a detailed customer profile combining sentiment analysis with behavioural insights derived from transaction data [25]. The deep learning architecture is designed to analyse and predict customer sentiments and purchasing behaviours, informed by the latest research [65]. This study is significant as it provides an innovative solution for predicting customer churn in e-commerce, aiding sustainability [45]. It also enables targeted retention strategies and personalized customer engagement [59]. Additionally, it contributes insights to big data analytics and customer relationship management in e-commerce, showcasing deep learning's potential in transforming business practices and enhancing customer experience [40].

This study examines the efficacy of Random Forest and XGBoost classifiers in conjunction with three upsampling techniques—SMOTE, ADASYN, and Gaussian noise upsampling (GNUS)—across datasets with varying class imbalance levels, ranging from moderate to extreme (15% to 1% churn rate). Employing metrics such as F1 score, ROC AUC, PR AUC, Matthews Correlation Coefficient (MCC), and Cohen’s Kappa, this research provides a comprehensive evaluation of classifier performance under different imbalance scenarios, focusing on applications in the telecommunications domain. The findings highlight that tuned XGBoost paired with SMOTE (Tuned_XGB_SMOTE) consistently achieves the highest F1 score and robust performance across all imbalance levels. SMOTE emerged as the most effective upsampling method, particularly when used with XGBoost, whereas Random Forest performed poorly under severe imbalance. ADASYN showed moderate effectiveness with XGBoost but underperformed with Random Forest, and GNUS produced inconsistent results. This study underscores the impact of data imbalance, with MCC, Kappa, and F1 scores fluctuating significantly, whereas ROC AUC and PR AUC remained relatively stable. Moreover, rigorous statistical analyses employing the Friedman test and Nemenyi post hoc comparisons confirmed that the observed improvements in F1 score, PR-AUC, Kappa, and MCC were statistically significant (p < 0.05), with Tuned_XGB_SMOTE significantly outperforming Tuned_RF_GNUS. While differences in ROC-AUC were not significant, the consistency of these results across multiple performance metrics underscores the reliability of our framework, offering a statistically validated and attractive solution for model selection in imbalanced classification scenarios.

Accurate job failure predictions in distributed cloud environments are vital for improving reliability, scheduling efficiency, and cost-effectiveness. Moreover, the dynamic variability and scale of production workloads make failure prediction a challenge to overcome. Although classical machine learning (ML) models, particularly Random Forests (RF) provide strong predictive performance, they can be computationally demanding under frequent retraining. Contrarily, Quantum Machine Learning (QML) methods leverage quantum-enhanced kernels to capture complex patterns, but standalone quantum models continue to have hardware limitations related and depict significant computational overhead. In this paper, we examine a hybrid classicalquantum ML framework for job failure prediction in large-scale distributed systems. A Hybrid Quantum Support Vector Machine (QSVM) is employed, blending classical RBF and quantum kernels, and benchmarked against RF. Experimental results show that <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\mathbf{R F}$</tex> attains high predictive accuracy (<tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\mathbf{A c c} \approx 0.98-0.99$</tex>, ROC AUC <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\approx 0.99$</tex>, AP <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\approx 0.98$</tex>), while hybrid QSVM achieves moderate accuracy <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$(\approx 0.80-0.82)$</tex>, but they exhibit distinct timing behavior. While the quantum kernel evaluation is computationally expensive, the fitting is faster when the kernels are precomputed. This aspect offers potential advantages in cloud environments where rapid retraining is needed under evolving workloads. Furthermore, our results emphasize that classical and quantum models are not competing but rather complementing each other. While classical models offer high accuracy, the hybrid quantum approaches provide efficiency trade-offs for large-scale, time-sensitive prediction tasks.

This work studies the feasibility of applying quantum kernel methods to a real consumer classification task in the NISQ regime. We present a hybrid pipeline that combines a quantum-kernel Support Vector Machine (Q-SVM) with a quantum feature extraction module (QFE), and benchmark it against classical and quantum baselines in simulation (hardware validation remains future work). Hyperparameters were selected via nested cross-validation on the training partition and then fixed for test evaluation; under these settings, the proposed Q-SVM attains 0.7790 accuracy, 0.7647 precision, 0.8609 recall, 0.8100 F1, and 0.83 ROC AUC, exhibiting higher sensitivity while maintaining competitive precision relative to classical SVM. All headline metrics are obtained via high-fidelity simulation. We interpret these results as an initial indicator and a concrete starting point for NISQ-era workflows and hardware integration, rather than a definitive benchmark. Methodologically, our design aligns with recent work that formalizes quantum–classical separations and verifies resources via XEB-style (Cross-Entropy Benchmarking) approaches, motivating shallow yet expressive quantum embeddings to achieve robust separability despite hardware noise constraints.

A model based on the random forest is constructed to repair the missing trade times in ETC transaction data. The driving speed and traffic volume characteristics of the vehicles in the ETC transaction data are analyzed, while the driving speed of the missed transaction vehicles, and the distance of the road section where they are located, are combined as input features to repair the missing transaction time. A one-day transaction data of a province is used to test. The analysis results show that the random forest model has a better restoration effect and has a smaller mean absolute error and root mean square error; its mean absolute error is 2.71 s, the highest accuracy among the compared methods, and the data are more accurate and complete after interpolation using the random forest model. This paper suggests that the research based on ETC transaction data should first adopt the processing method in this paper to repair the missing trade time in the transaction data to improve the integrity of the data used and ensure the validity and accuracy of the relevant calculation results.

Customer churn is a critical issue impacting enterprises and organizations, particularly in the emerging and highly competitive telecommunications industry. It is important to researchers and industry analysts interested in projecting customer behavior to separate churn from non-churn consumers. The fundamental incentive is a firm’s intent desire to keep current consumers, along with the exorbitant expense of gaining new ones. Many solutions have been developed to address customer churn prediction (CCP), such as rule-based and machine learning (ML) solutions. However, the issue of scalability and robustness of rule-based customer churn solutions is a critical drawback, while the imbalanced nature of churn datasets has a detrimental impact on the prediction efficacy of conventional ML techniques in CCP. As a result, in this study, we developed intelligent decision forest (DF) models for CCP in telecommunication. Specifically, we investigated the prediction performances of the logistic model tree (LMT), random forest (RF), and Functional Trees (FT) as DF models and enhanced DF (LMT, RF, and FT) models based on weighted soft voting and weighted stacking methods. Extensive experimentation was performed to ascertain the efficacy of the suggested DF models utilizing publicly accessible benchmark telecom CCP datasets. The suggested DF models efficiently distinguish churn from non-churn consumers in the presence of the class imbalance problem. In addition, when compared to baseline and existing ML-based CCP methods, comparative findings showed that the proposed DF models provided superior prediction performances and optimal solutions for CCP in the telecom industry. Hence, the development and deployment of DF-based models for CCP and applicable ML tasks are recommended.

Customer churn presents a big challenge in the industry. Businesses have to deal with the problem of customers stopping using their products and services due to dissatisfaction, competitive offers, more affordable alternatives, or changing needs. Churn can be damaging to businesses since it causes revenue loss and higher costs. To address this issue, our research aimed to develop a prediction model that helps predict customer churn. We started with getting the data set about telecommunication. Our analysis and model development were based on this dataset. Then we did data visualization to gain a better understanding of the data through multiple charts. After that, we performed data preparation. First, we did data transformation, data cleansing to address missing values and outliers; feature selection was done, and finally, in this step, the data set was split into testing and training sets. Multiple machine learning algorithms were used for modeling, such as decision trees, random forests, logistic regression, support vector machines, Naïve Bayes, and neural networks. Following model development, we evaluated the model performance with each algorithm using tune model hyperparameters. The decision tree algorithm performed the best with %96.7 accuracy, %96.9 precision, %99.3 recall, and %98.1 F1-score. These findings showed how effective decision tree algorithms are in predicting customer churn. This predictive model will enable telecommunication businesses to predict potential churn, make retention strategies, reduce customer churn and increase customer retention rates.