Data Science

Data Science project lifecycle

April 7, 2020

Hey, we are back to yet another exciting post you! So, you are looking for what a Data Science project lifecycle looks like. Well, this post is crafted just to address this. In this article, we will discuss The phases and the steps involved in a Data Science project. To understand this, we will proceed sequentially. By the end of this article, you’ll know;

  1. What Project Lifecycle means?
  2. Significance of understanding the project lifecycle in Data Science
  3. Stages and phases involved in a typical Data Science project
  4. Steps and processes involved at each phase of a Data Science project

Let us get started with first knowing what a ‘Project Lifecycle’ actually means. 

What does a Project lifecycle mean?

A ‘Project lifecycle’ simply means the stages involved in the execution of a project from its starting to closure. The word ‘project’ here can represent projects from any industry or domain. Typically, a project goes through four major ‘evolutionary’ phases. These are namely ‘Initiation,’ ‘Planning,’ ‘Execution,’ and ‘Closure.’ These phases can be further divided into specific ‘Stages’ through which a Project undergoes. In this article, our focus is on Data Science projects. We will now see the significance of ‘Project Lifecycle’ in context of Data Science.

What is Data Science project lifecycle?

The term ‘Data Science project Lifecycle’ means typical project evolution phases involved in executing a Data Science project. A data science project lifecycle starts from data collection/acquisition and ends at operations and optimization of the model. Let us proceed to see what these phases and stages are!

Phases and stages featured in a typical Data Science project

In this part, you will come across seven stages involved in executing a Data Science project. These seven stages lie under four phases of project progress timeline. Now let us see what they are and we will then discuss each of it in detail.

  1. Initiation phase
    • Data Acquisition stage
    • Data preparation stage
  1. Planning phase
    • Hypothesis and modelling stage
  1. Execution phase
    • Evaluation and interpretation stage
    • Deployment stage
    • Operations stage
  1. Closure phase  
    • Optimization stage

Initiation phase

The ‘initiation phase’ of a data science project marks the beginning of the project. It involves the following two stages;

#1. Data Acquisition stage

A data science project begins with sourcing and acquisition of the data. The type of data that needs to be sourced depends on the problem one is solving. For example, to forecast sales, you may need to source the historical sales data of a product and data on the ‘purchase trends’ of your customer base. Or, you might need historical records of disease outbreaks, weather data, geographical information, etc. to predict epidemic outbreaks. You can even utilize ‘Real-time’ data to make business decisions. Therefore, it becomes necessary to define the objective of the problem we are solving. So the following becomes the starting point of our project life-cycle. 

  • Defining the objective: The data scientists or engineers must pose questions like ‘what are model is going to do?’ and ‘What kind of data would we need for it to function?’ For example, your project might aim at predicting the weather so that you will need the weather data. Or, its objective could be to filter spam emails so that you will require large data samples of spam and non-spam emails. If your application is for a business case, then you should use the data accordingly. 

In essence, you need to figure out which type(s) of datasets you’ll need to solve the problem you’ve targeted. Now where do you source the data from? Here are some commonly used sources to harvest massive datasets.

  • Web server logs
  • Data harvested from social media platforms
  • Online repositories like government data (Census, healthcare, employment, economy, etc.)
  • Directly streamed data through APIs
  • Other sources like surveys and enterprise data, etc.

#2. Data Preparation stage

In this stage, we preprocess or prepare the data that we sourced in the first stage. Data preparation stage is also known as the ‘Data wrangling’ or ‘Data cleansing’ phase. This is a crucial step in the life-cycle of such a project. This is because the success of the project depends upon the quality of the data being fed to it. Often, the sourced data is not ready to be used right away. This is because it may contain wrong information, missing information or misplaced information. The acquired data may also need transformation, i.e., converting it into a usable form. 

It must be noted that this is a time-consuming stage in the data science project cycle because it includes exploratory analysis operations. Here are the processes involved in Data cleaning/preparation.

  • Replacing or removing the missing information/entries.
  • Correcting the semantic errors in the dataset.
  • Checking for possible biases in the data by tracking the origin source.
  • Transforming the data entries into useable format. For example, if the data column reflects the dates of highest sales, then it may not be useful. Therefore, the dates can be transformed into days, months and time to spot patterns and insights.

Another common step in this stage is setting-up a data-pipeline. A data-pipeline is necessary to feed the data regularly. This data is refreshed and must be aggregated regularly to keep the project model’s performance at its peak.

At the end of this stage, the collected data is ready to be loaded into the suitable tool/model for use. For that, data-warehouses are used. From a data-warehouse, an organization can borrow a structured dataset that can solve its problems. Arriving at the next phase.

Planning phase

As the name suggests, this phase involves theorizing and planning for the project model. In case of data science, the projects are based on algorithmic operations and Machine Learning models. The following is the stage involved in the planning phase of the data science lifecycle.

#3. Hypothesis and modelling stage

This stage is the combination of planning, theorizing, and execution of a data science project. It begins with theorizing a model to crunch the prepared data. We can either process the data to reveal insights via directly applying computational algorithms or build an ML model to do so. For example, previous year’s customer trends can be revealed via direct algorithmic operations, while analyzing customer trends in real-time will require a machine learning model. 

When the model is finalized, it is then time to build it, run it and refine it for the best output. This step involves coding a suitable program in MATLAB, Python, R or Perl. The model is then trained with a suitable learning algorithm (Like Classification and regression tree, Iterative Dichotomiser 3, Gaussian Naive bias, k-means,etc.). Two major steps are involved here.

  • Feature engineering: In this step, data attributes or features are created for training the data science model. For example, in a Census data of the population of New Delhi, ‘name,’ ‘age,’ and ‘gender’ represent the features for every individual entry ‘person.’ 
  • Model training: Once the features are defined, a suitable learning algorithm sweeps the cleaned data and learns from it (As given in the last paragraph). For example, a binary model would require a binary-classification algorithm and so on. To train the model, the available dataset is ‘split’ into a ‘training dataset’ and ‘test dataset.’ Usually, this split ratio is 70/30. This means, 70 percent of the data will train the learning algorithms and the rest 30 percent will be used to test its prediction score.

The model is then checked for bugs and errors. It is refined until it becomes usable as an application.

Execution phase

This is the phase where the action happens. Execution phase involves evaluation of the data science application built in the previous phase, deploying it over a suitable platform and then operating it for the thought-our purpose. Take a look at the stages involved in this phase.

#4. Evaluation of the model

In this stage, the Data Science model is evaluated by its performance. For evaluation, several metrics are judged to know how accurately a model is performing. These metrics and evaluation methods are different for different models. For example, if your model aims at predicting everyday stock values, then it will be evaluated by the Root Mean Square Error (RMSE) method. If your data science project involves a Binary-classification model, say for classifying spam emails, then Area Under the Curve (AUC) and Log-loss parameters will be judged. One important thing to note is that the model must be compared and evaluated using the validated and test datasets. This process is repeated until our project model starts giving accurate and actionable predictions.

#5. Deployment stage

This stage is where the tested and validated Data Science model is deployed on the web or a particular platform for use. Before doing so, the model needs to be ‘operationalized.’ Operationalizing the model is nothing but placing the model in a production-like environment, where it is supposed to perform. The models are supposed to make predictions on batches of data or real-time data. The advantage of testing the model in this way is that it can be tweaked as per the real-world scenario of deployment.

The project model is now ready to be deployed. The model is exposed to open API interface to deploy over the web. After doing so, the model/application can be consumed by websites, dashboards, spreadsheets, back-end applications and business applications. For example, you can deploy a Data Science ML model on Azure Cloud network after creating and testing it.

#6. Operations stage

This stage involves the development of a strategy for operating and maintaining a data science project for a long duration. Operational performance is majorly measured by monitoring the ‘Downgrade’ factor of the model. It is then tweaked and calibrated accordingly. Then the project maintenance plan comes into effect. Data science project maintenance usually involves the integration of monitoring and telemetry systems into the model. This enables the engineers and data scientists to gather project log reports and efficient troubleshooting. 

Once the above step is executed, the data science project application is fully functional and in use. The closure phase of the project life-cycle has arrived at this point.

Closure phase

This is the last phase of a data science project execution. In the closure phase, the model is up and running. It is already being consumed by various online applications through the open API interfaces. This phase is the end-point of a typical Data Science project.

#7. Optimization stage

Massive amounts of data are generated every day. Hence, there arises a need for the models to keep learning and training according to the newly generated data. The Data Science models need to adapt to this new information. And, enabling them to do so is called ‘Optimization.’ This stage is nothing but keeping up the model’s performance by taking necessary steps. 

There! We just came through the typical life-cycle of a Data Science project. You must know that the above-listed sequence and steps are just to give you an outline of how such a project is tracked. The stages involved can be altered at any point as per the project requirements. We hope that you now hold a clearer idea of the Data Science project life-cycle.

Also Read: How to learn data science with no coding background?