emmanuelkalunga / crisp-dm-assessment

An assessment for data scientist

CRISP-DM assessment

Description

The Cross Industry Standad Process for Data Mining (CRISP-DM) is a good practice in machine learning and is considered by many to be a wholistic and structured framework to solve machine learning problems. following such a process allows the data scientist to understand the problem, ask approriate questions, understand the data available for the analysis and trainig of an evantual model to describe the problem at hand. Following CRISP-DM process is thus hilghly recommanded. It consist of the following stages:

1. Business Understanding
2. Data Understanding
3. Data Preparation
4. Modeling
5. Evaluation
6. Deployment

The objective of this assessment is to demonstrate the ability to solve machine learning problems following the CRISP-DM stages, and making informed decisions with regards to the approaches taken for feature engineering and modelling.

As a use case, we will use a churning problem, where, based on historical churning data, a classification model should be designed and trained to predit customers churn in a telecommunicaton company.

Data

The data used (found in the data folder) is sampled from the churning dataset. Each row represents a customer, each column contains that customer’s attributes, and the target column "churn".
Part of the data is withheld for evaluation purposes.

Submission and evaluation

Submit the commented code (A python notebook would be ideal) including explanations on choices made. Also submit all necessary codes and files used for deplyement.

There is no right or wrong, the submission will be evaluated more on the general understanding of machine learning along the CRISP-MD process, than on the prediction accuracy (though a good ML will result in a adequate prediction accuracy) The main focus of the evaluation will be around the following points:

• Quality of the resulting algorithm (i.e. feature engineering and classification model):

An algorithm that demonstrates clear understanding of the the approach taken and the tools used, and results in good prediction evaluation metrics.

• Model deployment, accessibility and scalability:

Online accessibility and scalability of the model using a tool of your choice (e.g. flask, cloud computing platforms, etc).
Regardless of the approach tacken, the deployed product should be accessible by a third party, either via a standalone package or an API endpoint. The deployed product should take unlabled data and return the classification result.

Note:

Preferably work through all steps of the CRISP-DM process including the deployemnt. In case you are not familiar with deployment of machine learning algorithm, focus on the algortihm only. Similarly if you have a strong machine learning background, focus on the deployment architecure. In this case, a simple classification algorithm will suffice.