For this lab and in the next lessons we will use the dataset 'Healthcare For All' building a model to predict who will donate (TargetB) and then - how much they will give (TargetD) (will be used for lab on Friday). You will be using files_for_lab/categorical.csv, numerical.csv, and target.csv which can be found at this link.
link to data
You will need to download the data locally. Remember to add the files to your .gitignore.
You are revisiting the Healthcare for All Case Study. You are provided with this historical data about Donors and how much they donated. Your task is to build a machine learning model that will help the company identify people who are more likely to donate and then try to predict the donation amount.
In this lab, we will take a look at the degree of imbalance in the data and correct it using the techniques we learned in the class. You should fork and clone this Repo and begin a new Jupyter notebook.
Here are the steps to be followed (building a simple model without balancing the data):
Everyone is starting with the same cleaned data
Begin the Modeling here
- Look critically at the dtypes of numerical and categorical columns and make changes where appropriate.
- Concatenate numerical and categorical back together again for your X dataframe. Designate the TargetB as y.
- Split the data into a training set and a test set.
- Split further into train_num and train_cat. Also test_num and test_cat.
- Scale the features either by using MinMax Scaler or a Standard Scaler. (train_num, test_num)
- Encode the categorical features using One-Hot Encoding or Ordinal Encoding. (train_cat, test_cat)
- fit only on train data, transform both train and test
- again re-concatenate train_num and train_cat as X_train as well as test_num and test_cat as X_test
- Fit a logistic regression (classification) model on the training data.
- Check the accuracy on the test data.
Note: So far we have not balanced the data.
Managing imbalance in the dataset
- Check for the imbalance.
- Use the resampling strategies used in class for upsampling and downsampling to create a balance between the two classes.
- Each time fit the model and see how the accuracy of the model has changed.