People new to Machine Learning thinks that handling the Machine Learning project is the simple linear process of creating the model but its never is. The proper cycle must be followed for the successful completion of ML projects. We have to often go back and forth different phase. Different organization may follow different cycle but in this post I have tried to elaborate most commonly followed life cycle.
Machine Learning life cycle generally have following four phases :
1. Business understanding & Problem Discovery : It includes the time when we understand the context about the project and stakeholder. We come out with the clear and specific answer that our model has to answer. In this phase we should decide how the model is gonna be used evaluated. The phase generally includes:
· Defining the clear Objective
· Being precise with problem definition
· Know the stakeholders well and their goals
· Know about the Data Sources and Development Environment
· Compare the existing practice with the project goals
2. Data Acquisition & Understanding : This phase is about gathering the necessary data and making sure that the data is in the form that it can be used to build the model. We should make sure that whether the current available data is enough to answer the question. This phase includes:
· Acquiring the data from relevant source
· Cleaning the data to remove the unnecessary Noise
· Building Data Pipeline
· Exploratory Data Analysis
· Feature Engineering to get more precise answer
3. Modelling and Evaluation : After we have cleaned and feature engineered the data we can start building the model. The most appropriate algorithm should be selected and test the model. We should repeat the process lot of time to make sure that we get a good result. If we didn’t get the good result we must modify, refine, iterate over the previous phases and test again. The phase generally includes:
· Model Training
· Model Selection
· Model Reporting
4. Delivering the Product : If we are happy with the product we have build we can prepare to deliver the product. This is very wrong assumption that building the good model completes all the task. The product must me delivered with the proper documentation of of each step. Stakeholders must me made aware of all the steps. The phase generally includes :
· Model Solution
· Documentation
· Knowledge Transfer
Test Case
Consider the following scenario , Diva owns a Agricultural Consultancy Firm and she wants the predict the yield that she will get when she uses X amount of fertilizer. She wants to use the fertilizer efficiently as possible both to decrease cost and environmental impact of the fertilizer.
Diva’s objective is to predict the yield. She identifies the data source of soil fertility, humidity, fertility and output for the past years. Diva must think whether the model she is gonna build can be used widely as a part of consultation firm.
Now, Diva should start gathering the data from relevant sources. She must be very careful handling the data and must make sure that the data is relevant and can solve the question. Data must be cleaned and should be analyzed with utmost care. She should build the proper report of her analysis because this will help her to convince her customer.
After data analysis and cleaning she can now start building her model. In this case it is a regression task so she should find most efficient regression algorithm. She should start with simple algorithm like Linear regression and then go for more complex ones. Once she is happy after the proper evaluation of the model she can move to the fourth phase and deploy the model and can use that for her organization.
