The project provides a ranking algorithm to predict the probability of purchase. In other words, when a potential customer comes in, we predict the chance of selling the product to the new customer. We rank the new customers according to the predicted probability and first assign to consultants the customers with a higher probability of purchasing the product.

• Python 3.11.7

1. Install required packages

   pip install -r requirements.txt
   

1. Change the directory to /src

   cd src
   

2. Run the following command to start the program and rank the potential customers based on predictions:

   python main.py 
   

By default, the following configurations are set:

model_type: logistic_regression # Model to be used for predictions
lasso_penalty: 10 # Regularization strength. It can accept a list of numbers
file_name: Customers_Dataset # Name of dataset file. The file should be placed in /src/datasets
n_estimators: 50 # Number of trees in random forest. It can accept a list of numbers. This value will be ignored when logistic regression is used.
show_plots: False # If plots should be displayed while running the program.
number_of_potential_customers: 20 # Number of samples against which rankings will be generated
sparse_column_threshold: 60 # Removes features which are empty based on this threshold

To run the program using random forest model, the following command can be used:

python main.py --model_type random_forest

The program uses the dataset from the /src/datasets directory. The default dataset filename is "Customers_Dataset.csv" which contains about 85k customers.

If a new dataset needs to be used, the csv file should be placed in the mentioned directory and the filename should be passed as an argument while running the program e.g. If the filename is new_customers_dataset.csv then the following command should be used

python main.py --file_name new_customers_dataset

The plots are automatically saved in /src/plots directory when the program runs.

The models are saved in the /src/models directory with .pkl extension.

The customers are ranked and displayed as an output in the terminal when the program runs and also gets saved in a csv format file in /src/results directory.

The docker image can be built using the Dockerfile to run the program. Run the following command to build the image:

docker build -t purchase-predictor .

The image will be built with the name "purchase-predictor". Afterwards, the following command can be used to run the docker container.

docker run purchase-predictor

To use different configuration while running the docker container, the arguments can be passed as shown below:

docker run purchase-predictor --sparse_column_threshold 70 --number_of_potential_customers 20

The Jupyter Notebook with the filename "exploratory_data_analysis.ipynb" contains the initial data exploration. It is placed in the /src directory.