anityagan9urde / hr-analytics-will-someone-quit

This is a Flask API, deployed on Heroku, that takes in employee info and predicts whether they will quit the job using an SVM model.

HR-Analytics: Predict whether someone will quit the job

• This is a Flask API developed by me to determine if a Data Scientist would leave their current job provided their previous information.
• Contents:
  • Dataset
  • Model Used
  • API
  • Deployment
  • What did I learn

Deploy the API on Heroku by clicking the button below.

Running the project locally:

- Clone this repository. Open CMD. Ensure that you are in the project home directory. Create the machine learning model by running SVM.py as such:

python SVM.py

- This will create a serialized version of our model into a file svm_model.pkl or use the pretrained model saved with the same name.

- Now, run Employee_api.py using below command to start Flask API

python Employee_api.py

- Open any browser and paste this URL: http://localhost:5000 to run the file as an app.

---

Following images show how the API will look when run properly:

---

---

Dataset:

• The dataset used for training the SVM model was taken from Kaggle.
• Link: HR Analytics: Job Change of Data Scientists: https://www.kaggle.com/arashnic/hr-analytics-job-change-of-data-scientists
• Features:
  • enrollee_id: Unique ID of the candidate
  • city: City code
  • city_development_index: Developement index of the city (scaled)
  • gender: Gender of the candidate
  • relevent_experience: Relevant experience of the candidate
  • enrolled_university: Type of University course enrolled (if any)
  • education_level: Education level of the candidate
  • major_discipline: Education major discipline of the candidate
  • experience: Candidate total experience in years
  • company_size: No. of employees in current employer's company
  • company_type: Type of the current employer
  • last_new_job: Difference in years between previous job and current job
  • training_hours: Training hours completed
• Target:
  • 0 : Not looking for job change,
  • 1 : Looking for a job change
    

Model Used:

• The dataset shows that this is clearly a classification task and can be solved by a myriad of classification algorithms such as Logistic Regression, Decision Trees and even Random Forests.
• I chose Support Vector Machines(SVMs) because of the flexibility it shows during training.
• After implementing Cross Validation using StratifiedKFold and doing parameter search using Grid Search CV on both SVM and Random Forest, I found out that the SVM performed slightly better in understanding the correlation between the features and the target.
• Following is the table for their individual scores.

Scoring Parameter	Random Forest	SVM
Accuracy	0.77	0.77
Precision(wgt)	0.75	0.75
Recall(wgt)	0.77	0.77
F1-score(wgt)	0.75	0.76

• Recall for each class:

Classes	Random Forest	SVM
0	0.92	0.90
1	0.32	0.38

• Hyper Parameters chosen for:
  • SVM: {'C': 411, 'kernel': 'rbf'}
  • Random Forest: {'criterion': 'entropy', 'max_depth': 9, 'max_features': 'sqrt', 'n_estimators': 425}
• Provided the dataset was slightly unbalanced, the SVM model gave a better Recall score for the negative classes as compared to Random Forest.
• Hence, I chose SVM as the model to use for the API.
  

API:

• I have made an API for the SVM model so that users can interact and use the model with ease.
• To make the API work I have used the Flask library which are mostly used for such tasks.
• I have also connected a HTML form to the flask app to take in user input and a CSS file to decorate it.
  

Deployment:

• The Flask API was deployed on the Heroku cloud platform so that anyone with the link to the app can access it online.
• I have connected this GitHub repository to Heroku so that it can be run on the Heroku dyno.
• I have used the Gunicorn package which lets Python applications run on any web server. The Procfile and requirements.txt should be defined with all the details required before the deployment.
  

What did I learn:

• Data Wrangling using Pandas
• Feature Engineering to fit our data to our model
• Selecting the right model using cross-validation
• Hyperparameter tuning
• Unbalanced datasets are hard to work with and finding the right scoring method
• Saving the model and using it again with Pickle
• Making a flask app
• A little frontend web development
• Making the app live by deploying it on cloud platforms

Back to Top