TY - GEN
T1 - Customer Churn Prediction in the Telecom Industry Via a Clustering-Random Forest-XGBoost Approach
T2 - Workshops of 26th International Conference on Computational Science and its Applications, ICCSA 2026
AU - Lee, Chien Sing
AU - Nanda, Sharini Sivadas
N1 - Publisher Copyright:
© The Author(s), under exclusive license to Springer Nature Switzerland AG 2027.
PY - 2027
Y1 - 2027
N2 - This study is framed within Technology Acceptance Models 1, 2, 3, in identifying key features which influence churn in the telecommunications industry. Data is obtained from 225 Malaysian survey respondents via random sampling. Mining adapts from Bach et al.’s 3-stage approach. For Phase 1, k-means, k-medoids and hierarchical agglomerative clustering are utilized to determine the most suitable clustering technique to segment customers. Subsequently, bar graphs form more detailed preliminary inferences and cluster profiling. In Phase 2a, features selected via correlation analysis are complemented with principal component analysis. In Phase 2b, non-linear relationships are assessed using Random Forest and XGBoost. The top 15 identified key features are ranked and compared. Outcomes are compared with findings in Phase I and Phase 2a. Findings highlight the viability of a combination of these techniques, depending on the research objectives, the nature of the dataset, the objectives/ nature of the respective algorithms and the domain investigated. Correlation (covariance matrix) and PCA loadings in Phyton reveal different polarities. Moreover, the delta value of PCA loadings are consistent across all clusters. Chi squared and p in CHAID performs comparably with RF and XGBoost. A very humble analogy to quantum states due to polarity, robustness in correlational feature selection reflecting information gain and communication’s parity bits to ensure data integrity in interconnected systems due to covariance-PCA loadings’ delta as identifiers, may be useful for STEM education, depending on model fit, and to finance, e.g. Buy Now Pay Later.
AB - This study is framed within Technology Acceptance Models 1, 2, 3, in identifying key features which influence churn in the telecommunications industry. Data is obtained from 225 Malaysian survey respondents via random sampling. Mining adapts from Bach et al.’s 3-stage approach. For Phase 1, k-means, k-medoids and hierarchical agglomerative clustering are utilized to determine the most suitable clustering technique to segment customers. Subsequently, bar graphs form more detailed preliminary inferences and cluster profiling. In Phase 2a, features selected via correlation analysis are complemented with principal component analysis. In Phase 2b, non-linear relationships are assessed using Random Forest and XGBoost. The top 15 identified key features are ranked and compared. Outcomes are compared with findings in Phase I and Phase 2a. Findings highlight the viability of a combination of these techniques, depending on the research objectives, the nature of the dataset, the objectives/ nature of the respective algorithms and the domain investigated. Correlation (covariance matrix) and PCA loadings in Phyton reveal different polarities. Moreover, the delta value of PCA loadings are consistent across all clusters. Chi squared and p in CHAID performs comparably with RF and XGBoost. A very humble analogy to quantum states due to polarity, robustness in correlational feature selection reflecting information gain and communication’s parity bits to ensure data integrity in interconnected systems due to covariance-PCA loadings’ delta as identifiers, may be useful for STEM education, depending on model fit, and to finance, e.g. Buy Now Pay Later.
KW - clustering
KW - Customer Churn Prediction
KW - delta
KW - Information Science
KW - PCA loadings
KW - Random Forest
KW - Telecom
KW - XGBoost
UR - https://www.scopus.com/pages/publications/105043993993
U2 - 10.1007/978-3-032-30518-3_19
DO - 10.1007/978-3-032-30518-3_19
M3 - Conference contribution
AN - SCOPUS:105043993993
SN - 9783032305176
T3 - Lecture Notes in Computer Science
SP - 289
EP - 307
BT - Computational Science and Its Applications – ICCSA 2026 Workshops - Proceedings
A2 - Gervasi, Osvaldo
A2 - Faginas Lago, M. Noelia
A2 - Murgante, Beniamino
A2 - Garau, Chiara
A2 - Rocha, Ana Maria A. C.
A2 - Braga, Ana C.
A2 - Karaca, Yeliz
PB - Springer Science and Business Media Deutschland GmbH
Y2 - 30 June 2026 through 3 July 2026
ER -