TY - GEN
T1 - Feature Weighting using Gravitational Search Algorithm in Customer Churn Prediction
AU - Hendro, Hendro
AU - Shiddiqi, Ary Mazharuddin
AU - Saikhu, Ahmad
N1 - Publisher Copyright:
© 2024 ACM.
PY - 2024/4/12
Y1 - 2024/4/12
N2 - Churn prediction is a critical issue for telecommunications companies due to the high customer turnover. Companies implement different retention strategies in response to this challenge, including predicting customer churn. Developing a model for predicting customer churn faces difficulties due to the complex nature of the data, particularly its high dimensionality, which complicates data analysis. Techniques for reducing the number of features can be implemented to enhance accuracy and reduce processing time. However, these methods carry the risk of losing essential information. Feature weighting techniques can be utilized to avoid it, wherein the weight assigned to a feature signifies its significance in the modelling or data analysis. This study applies the Gravitational Search Algorithm to assign weights to features automatically using KNN classification. The investigation is carried out on a publicly available dataset, comparing the performance of a baseline model without feature weighting to a model that incorporates feature weighting. The findings validate that the proposed model achieves an accuracy of 89%, indicating its effectiveness in improving accuracy compared to the baseline method. This research also compares our results with others' research and shows that our model has some improvements in several metrics.
AB - Churn prediction is a critical issue for telecommunications companies due to the high customer turnover. Companies implement different retention strategies in response to this challenge, including predicting customer churn. Developing a model for predicting customer churn faces difficulties due to the complex nature of the data, particularly its high dimensionality, which complicates data analysis. Techniques for reducing the number of features can be implemented to enhance accuracy and reduce processing time. However, these methods carry the risk of losing essential information. Feature weighting techniques can be utilized to avoid it, wherein the weight assigned to a feature signifies its significance in the modelling or data analysis. This study applies the Gravitational Search Algorithm to assign weights to features automatically using KNN classification. The investigation is carried out on a publicly available dataset, comparing the performance of a baseline model without feature weighting to a model that incorporates feature weighting. The findings validate that the proposed model achieves an accuracy of 89%, indicating its effectiveness in improving accuracy compared to the baseline method. This research also compares our results with others' research and shows that our model has some improvements in several metrics.
KW - customer churn prediction
KW - feature weighting
KW - gravitational search algorithm
UR - http://www.scopus.com/inward/record.url?scp=85197265510&partnerID=8YFLogxK
U2 - 10.1145/3661725.3661787
DO - 10.1145/3661725.3661787
M3 - Conference contribution
AN - SCOPUS:85197265510
T3 - ACM International Conference Proceeding Series
BT - CMLDS 2024 - 2024 International Conference on Computing, Machine Learning and Data Science, Conference Proceedings
PB - Association for Computing Machinery
T2 - 2024 International Conference on Computing, Machine Learning and Data Science, CMLDS 2024
Y2 - 12 April 2024 through 14 April 2024
ER -