Personalized Customer Churn Analysis with Long Short-Term Memory

Bayrak, Ahmet Tuğrul; Aktas, Asmin Alev; Tunalı, Okan; Susuz, Orkun; Abbak, Nese

doi:10.1109/bigcomp51126.2021.00024

Cited by 5 publications

(6 citation statements)

References 7 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…In the second step, data is organized and analyzed to explore patterns and relationships. First, a threshold value should be de ned to distinguish "churn customer" from "ordinary customer" [17]. The time frame when several customers are inactive is also called the "time window".…”

Section: Customer Churn Analysismentioning

confidence: 99%

“…Afterward, training data is prepared for the data mining process. [17]. As the training data is tested and veri ed, it can be used as input for a predictive model [8].…”

Section: Customer Churn Analysismentioning

confidence: 99%

“…A predictive model for churn analysis is built by following ve steps [19]: The rst step is collecting appropriate data for analysis. Demographic data, customer account information, call data in the telecommunication sector [19], and sales data in the retail sector [17] are examples of data used in CCA. In the second step, data is organized and analyzed to explore patterns and relationships.…”

Section: Customer Churn Analysismentioning

confidence: 99%

See 2 more Smart Citations

A novel robust LSTM model on Churn and comparative analysis versus deep learning, machine learning and ensemble learning: Telecommunications Case

Başarslan,

ÜNAL,

KAYAALP

2023

Preprint

View full text Add to dashboard Cite

Customer churn is an important issue in increasing both the long and short-term revenues of a company. If companies identify customers’ churn behavior, they can prevent customer loss, ensure customer loyalty, and in turn gain better financial returns. The telecommunication sector is a customer-oriented sector that requires customer retention to survive in the market and, in this sector customer loss is observed at a high level. In recent years, artificial intelligence-based customer churn analysis has been widely used to predict customers’ churn behavior. In this study, a customer churn analysis was conducted with publicly shared Telco telecom data. Predictive models were built with machine learning (LR, KNN, SVM, DT, RF, ANN), ensemble learning (XGBoost, Voting), and deep learning (LSTM) methods. Besides, a 3-layered LSTM model was proposed. Accuracy, F1, precision, and recall rates were used to evaluate the models. As a result, the novel 3-layered LSTM model achieved 91% accuracy, 87% precision, 84% recall, and 89% F1 values. The proposed model is competitive with the literature.

show abstract

Section: Customer Churn Analysismentioning

confidence: 99%

“…Afterward, training data is prepared for the data mining process. [17]. As the training data is tested and veri ed, it can be used as input for a predictive model [8].…”

Section: Customer Churn Analysismentioning

confidence: 99%

Section: Customer Churn Analysismentioning

confidence: 99%

See 1 more Smart Citation

A novel robust LSTM model on Churn and comparative analysis versus deep learning, machine learning and ensemble learning: Telecommunications Case

Başarslan,

ÜNAL,

KAYAALP

2023

Preprint

View full text Add to dashboard Cite

show abstract

“…Bayrak et al [30] discussed that churn prediction is critical to achieving. It's not easy to predict churning clients.…”

Section: Literature Survey Ahmad Et Almentioning

confidence: 99%

An Efficient Hybrid Classifier Model for Customer Churn Prediction

M A,

K K

2022

International Journal of Electronics and Telecommunications

View full text Add to dashboard Cite

Customer churn prediction is used to retain customers at the highest risk of churn by proactively engaging with them. Many machine learning-based data mining approaches have been previously used to predict client churn. Although, single model classifiers increase the scattering of prediction with a low model performance which degrades reliability of the model. Hence, Bag of learners based Classification is used in which learners with high performance are selected to estimate wrongly and correctly classified instances thereby increasing the robustness of model performance. Furthermore, loss of interpretability in the model during prediction leads to insufficient prediction accuracy. Hence, an Associative classifier with Apriori Algorithm is introduced as a booster that integrates classification and association rule mining to build a strong classification model in which frequent items are obtained using Apriori Algorithm. Also, accurate prediction is provided by testing wrongly classified instances from the bagging phase using generated rules in an associative classifier. The proposed models are then simulated in Python platform and the results achieved high accuracy, ROC score, precision, specificity, F-measure, and recall.

show abstract

“…Considering the presented scenarios and the two drawbacks of the commonly used approaches, namely, (1) the inabil- Using an individualized value solves the issue of not fitting every player's behavior (solving problem 2). This labeling approach was applied in the fast-food industry by Bayrak and colleagues (Bayrak et al 2021) to personalize the churn prevention system according to each customer's behaviors.…”

Section: Individualized Fixed Valuementioning

confidence: 99%

Optimizing Profit by Mitigating Recurrent Churn Labeling Issues: Analysis from the Game Domain

Humenhuk

Kummer

Paraíso

2021

AIIDE

View full text Add to dashboard Cite

Churn can be interpreted as customer defection and can be considered one of the most critical challenges in the Game Analytics domain because of its impact on the game industry's profit. When predicting churn, the first step is defining what is considered churn, which can change depending on the players' behaviors and approaches. This work studied related works and revealed two recurrent issues in the labeling process: limitations on the adopted labeling approaches (1) and the static definition of churn (2). To mitigate the first issue, an individualized labeling approach was deployed. To address the second one, a novel evaluation method, based on the impact of a change in the churn definition, was proposed. This method allowed the proposition of two new labeling approaches, which were included in the analysis. By comparing the labeling approaches in two games using a profit perspective, it was identified that the new ones present statistically significant benefits compared to the traditional ones. Regarding the evaluation method, its usage can justify when the redefinition of churn and the classifier's retraining should happen to improve profit. The results are valuable for the game context, potentially extended to other contexts by delivering more reliable labels and more validated classification performance.

show abstract

Personalized Customer Churn Analysis with Long Short-Term Memory

Cited by 5 publications

References 7 publications

A novel robust LSTM model on Churn and comparative analysis versus deep learning, machine learning and ensemble learning: Telecommunications Case

A novel robust LSTM model on Churn and comparative analysis versus deep learning, machine learning and ensemble learning: Telecommunications Case

An Efficient Hybrid Classifier Model for Customer Churn Prediction

Optimizing Profit by Mitigating Recurrent Churn Labeling Issues: Analysis from the Game Domain

Contact Info

Product

Resources

About