a. Full name (as in NRIC) and email address (stated in your application form). Name: TONG Shanshan, Email. tongshanshan117@gmail.com
b. Overview of the submitted folder and the folder structure.
Data src download.py process-data.py model.py eda.ipynb README.md requirements.txt run.sh
c. Instructions for executing the pipeline and modifying any parameters click the ‘run.sh’ file to run the .py files
d. Description of logical steps/flow of the pipeline. If you find it useful, please feel free to include suitable visualization aids (eg, flow charts) within the README. download.py : download the database process-data.py : preprocess the data including data cleaning, feature engineering, data standardization model.py: train and test the models
e. Overview of key findings from the EDA conducted in Task 1 and the choices made in the pipeline based on these findings, particularly any feature engineering. Please keep the details of the EDA in the .ipynb. The information in the README.md should be a quick summary of the details from .ipynb.
The key findings from the EDA include:
- 17% of club members opt for attrition.
- On average, members who opt for attrition spend 2 hours less than those who keep membership.
- Thomson Branch with most members has the lowest attrition rate of 15%, which is 5% less than other two branches (Changi and Krangi). For female members, Thomson has the lowest (12%) attrition rate while Krangi has the highest (30%).
- Members with travel time > 45 mins show a higher Attrition Rate (21%) than those with Travel time < 45 mins (15%-16%).
- Members holding higher qualifications and lower tier membership tends to have lower attrition rates. This trend is more evident among female members.
f. Explanation of your choice of models for each machine learning task. 4 models are selected to predict attrition, they are: Logistic Model, Random Forest Model, LightGBM Model and SGDClassifier Model.
g. Evaluation of the models developed. Any metrics used in the evaluation should also be explained. Targets 0 1 Model Precision Recall f1-score Precision Recall f1-score Accuracy Logistic 0.84 0.71 0.77 0.22 0.38 0.28 0.65 Random Forest 0.85 0.92 0.88 0.38 0.23 0.29 0.80 LightGBM 0.84 0.93 0.89 0.39 0.20 0.22 0.80 SGDClassifier 0.84 0.93 0.89 0.39 0.20 0.27 0.80
h. Other considerations for deploying the models developed. in the future, i will create a m/flow and deploy it