Objective: To predict the customer churn for the e-commerce business and identify the main features affected to the customer churn.
SAS e-Miner: This tool used to the process for Sample, Explore, Model and Assess in SEMMA.
Talend Data Intergration: This tool used to combine the dataset.
Talend Data Prep: This tool used to process the data.
Dataset1.csv : First dataset Dataset2.csv : Second dataset E Commerce Dataset.xlsx : Combined dataset em_save_TRAIN_Latest Preparation : Processed dataset
- The dataset is self generated and refer to this link as reference : https://www.kaggle.com/datasets/ankitverma2010/ecommerce-customer-churn-analysis-and-prediction/data
dataprep_em_save_TRAIN_Latest Preparation_2024-01-07_11-45-40-773 : Code export from the Talend Data Prep of data processing
SAS e-Miner: Import the dataset into the SAS Enterprise Miner and add the required node for the process to carry out SEMMA
Talend Data Intergration: Import the two dataset into it and use the tMap to combine it and export using the tFileOutputDelimiter.
Talent Data Prep: Upload the combined dataset into it and perform data cleaning.
Gradient Boosting has outperformed compare to the Decison Tree and Random Forest.
| Decision Tree | Random Forest | Gradient Boosting | |
|---|---|---|---|
| Accuracy | 0.77 | 0.79 | 0.82 |
| TPR (Recall) | 62 / (62+14) = 0.816 | 62 / (62+14) = 0.816 | 62 / (62+14) = 0.816 |
| TNR | 56 / (56+20) = 0.737 | 58 / (58+18) = 0.763 | 62 / (62+14) = 0.816 |
The challenges that faced in this project is the data imbalanced problem. By applying with the data oversampling method, it has successfully improved the performance of model in predicting the churn.