Scribd
Upload
6 views6 pages

Neural Network Churn Prediction Analysis

The document outlines a Data Science assignment completed by a group of four members, detailing their use of the Iranian churn dataset from UC Irvine. It evaluates the performance of three classifiers, concluding that the Neural Network model outperforms others with the highest evaluation metrics, including AUC and accuracy. Additionally, an extra dataset was created for forecasting, utilizing the Neural Network model to predict missing churn data.

Uploaded by

hoanglongg2203
Copyright
© All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as DOCX, PDF, TXT or read online on Scribd
6 views6 pages

Neural Network Churn Prediction Analysis

The document outlines a Data Science assignment completed by a group of four members, detailing their use of the Iranian churn dataset from UC Irvine. It evaluates the performance of three classifiers, concluding that the Neural Network model outperforms others with the highest evaluation metrics, including AUC and accuracy. Additionally, an extra dataset was created for forecasting, utilizing the Neural Network model to predict missing churn data.

Uploaded by

hoanglongg2203
Copyright
© All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as DOCX, PDF, TXT or read online on Scribd

Data Science assignment

Group members

STT THÀNH VIÊN MSSV

1 Nguyễn Khánh Duy 31241021157

2 Hà Nữ Tường Vi 31241020690

3 Nguyễn Hữu Minh Tuấn 31241021140

4 Nguyễn Hoàng Long 31241024111

I downloaded a dataset from UC Irvine via following link :


[Link]
Forecast_date link:
[Link]
jtXt-i5FeuMWs/edit?gid=1609696779#gid=1609696779
Part1.
Part2.
Based on the evaluation metrics, the Neural Network model performs the best
among the three classifiers. It achieves the highest AUC (0.985), classification
accuracy (0.954), F1 score (0.954), precision (0.953), recall (0.954), and MCC
(0.824), indicating that it provides both strong discriminative power and balanced
performance across classes. The confusion matrix also shows that the Neural
Network correctly classifies the majority of both positive and negative instances,
with relatively few misclassifications. With only 82 FN errors, the Neural Network
dramatically minimizes the number of customers who will leave but are incorrectly
predicted to stay, a significant improvement over the Logistic Regression's 269 FN
errors. In contrast, the SVM model performs substantially worse, particularly with
a low AUC of 0.433 and lower precision and recall, suggesting poor separation
between classes. Logistic Regression performs moderately well but still falls short
compared to the Neural Network in nearly every metric. Therefore, the Neural
Network is clearly the most effective model for this dataset, demonstrating superior
accuracy, robustness, and predictive reliability.
From that, we also created an additional dataset for forecasting, in which the churn
column was missing, and we used the Neural Network model to predict it.

You might also like