Naslov
Late vs. early churn prediction and determinants

Autori
Perišić, Ana ; Šarlija, Anđela

Vrsta, podvrsta i kategorija rada
Sažeci sa skupova, sažetak, znanstveni

Izvornik
Book of Abstracts 19th International Conference on Operational Research / Mijač, Tea ; Šestanović, Tea - Zagreb, 2022, 79-79

Skup
9th International Conference on Operational Research (KOI 2022)

Mjesto i datum
Šibenik, Hrvatska, 28.09.2022. - 30.09.2022

Vrsta sudjelovanja
Predavanje

Vrsta recenzije
Međunarodna recenzija

Ključne riječi
churn prediction, late churn, early churn, churn determinants

Sažetak
Identifying churners became a core strategy to survive for businesses in various industries. This process implies selecting relevant features and building the churn prediction model, which is highly dependent on the churn definition statement and associated churn window size. Short churn window size is related to early churn prediction modeling, while late churn allows for long churn window size. Too short windows might mislabel customers as churned, leading to a high false- positive rate of churn, while too long windows may result in irreversible loss of customers. The duration of the churn window size may vary depending on different business goals. The main goal of this work is to examine the dependency of churn prediction models on the churn window size. We focus on comparing prediction performance and feature importance with respect to churn window size. The proposed methodology starts with a fixed set of features related to customer behavior data, which is followed by building separate churn prediction models for different churn window sizes. The predictive performance of churn prediction models and feature importance are assessed depending on the churn window size.

Izvorni jezik
Engleski

Znanstvena područja
Matematika, Interdisciplinarne prirodne znanosti, Ekonomija, Interdisciplinarne društvene znanosti

POVEZANOST RADA