THE DATA SCIENCE PROCESS

A Complete & Simple Guide

Based on Unit 2 — PVP Siddhartha Institute of Technology, Dept. of IT

Covering All Six Steps in Detail:

Step 1: Setting the Research Goal

Step 2: Retrieving Data
Step 3: Data Preparation
Step 4: Data Exploration
Step 5: Data Modeling
Step 6: Presentation & Automation
Introduction: What is the Data Science Process?

Data science is not just about writing complicated code or building powerful machine
learning models. At its heart, data science is about answering questions using data. It is
about taking raw, messy, unorganized information and turning it into useful knowledge
that helps people and businesses make better decisions.

However, getting from 'raw data' to 'useful knowledge' is not always simple. There are
many steps involved, and if you skip any one of them or do them in the wrong order,
your entire project can fail. This is why data scientists follow a structured process — a
clear step-by-step plan that helps them stay organized and produce reliable results.

The data science process taught in Unit 2 consists of six major steps. These steps are
not just theoretical concepts — they are practical actions that every real-world data
scientist performs in every project. Whether you are working at a small startup or a large
multinational company, these steps are the foundation of successful data science work.

Why Follow a Process?

Following a structured process helps you maximize chances of success at the lowest cost. It
also allows teams to work together efficiently, with each member contributing their expertise
to the right stage of the project. Without a process, projects become chaotic, results become
unreliable, and time and money get wasted.

Think of the data science process like building a house. You would not start painting the
walls before laying the foundation. Similarly, in data science, you cannot start building
models before understanding what you want to predict and gathering the right data.
Each step builds on the previous one.

In this document, we will walk through all six steps of the data science process in detail,
using simple language and everyday examples to make each concept clear and easy to
understand. We will also discuss an important idea: while the process is useful, you
should not follow it blindly. Real projects require flexibility and judgment.
Step 1: Setting the Research Goal

The very first step in any data science project is to figure out exactly what you are trying
to do. This sounds simple, but it is actually one of the most important and most
commonly skipped steps in data science. Many projects fail not because the data was
bad or the model was wrong, but because nobody took the time to clearly define what
the project was supposed to achieve.

Understanding the What, Why, and How

Before you touch a single piece of data, you need to answer three fundamental
questions:

What: What does the business or organization expect you to do? What is the specific
question you are trying to answer? For example, 'Which customers are most likely to
stop using our service in the next three months?'
Why: Why does this question matter? Why is the business interested in this topic?
Understanding the motivation behind the project helps you focus on what truly matters.
Maybe the company is losing customers and needs to act fast.
How: How will the results be used? Will the model be used to send targeted emails?
Will it be shown in a dashboard to managers? Knowing how the results will be used
affects what kind of model you build.

Answering these three questions ensures that you, your team, and the business are all
on the same page from day one. It prevents the frustrating situation where you spend
months doing research, only to present your results and have everyone say, 'That is not
what we wanted.'

Real-World Example
Imagine a mobile phone company asks you to 'analyze customer data.' That is too vague.
After asking the right questions, you discover they want to predict which customers will
cancel their subscription next month, so the sales team can call them and offer a discount.
Now you have a clear, specific goal that guides everything else.

The Project Charter

Once you have a clear understanding of the goal, the next task is to write a project
charter. A project charter is like a formal agreement or contract between you and your
client. It documents everything that was discussed and agreed upon at the start of the
project.

A good project charter includes the following elements:

1. A clear research goal — a one or two sentence statement that describes exactly
what the project aims to achieve.
2. The project mission and context — background information explaining why this
project is important and how it fits into the bigger picture of the organization.
3. How the analysis will be performed — a brief description of the methods, tools,
and techniques you plan to use.
4. What resources are needed — data sources, software tools, team members, and
time required.
5. Proof of concept — evidence that the project is feasible and has been done
successfully in similar situations.
6. Deliverables and success measures — exactly what you will produce and how
success will be measured.
7. A timeline — a realistic schedule showing when each phase of the project will be
completed.

The project charter protects everyone involved. Clients know what they are paying for.
Data scientists know what is expected of them. Team members know their roles and
responsibilities. It removes ambiguity and sets up the project for success.

Common Mistakes to Avoid

Many experienced data scientists admit that this step is where most failures begin. Here
are common mistakes to avoid:

• Rushing through this phase because you are eager to start coding or analyzing
data.
• Accepting a vague goal without asking follow-up questions.
• Assuming you understand the business problem without verifying your
understanding with the client.
• Not documenting the agreed-upon goal, leading to confusion and disagreements
later.
The golden rule of this step is: never assume. Always ask, verify, and document. The
time you invest here will save you from costly mistakes later in the project.
Step 2: Retrieving Data

Once you know exactly what you are trying to achieve, the next step is to gather the
data you need to achieve it. Data retrieval is the process of finding, accessing, and
collecting relevant data from various sources. This step sounds straightforward, but it
can be surprisingly complex in practice.

Data comes in many forms. It could be stored in a company database, scattered across
Excel files, available through an online service, or it may need to be purchased from a
third-party provider. Your job is to find the right data, in the right format, and make sure
it is of good enough quality to be useful.

Key Insight
Data is often compared to a diamond in the rough. In its raw form, it might look messy and
confusing. But with the right preparation and polishing, it becomes extremely valuable. The
data retrieval step is where you begin the process of finding and extracting those diamonds.

Start Inside Your Organization

The first place to look for data is within your own organization. Most companies collect
and store enormous amounts of data as part of their daily operations. Before looking
elsewhere, check if this internal data can meet your needs.

Internal data is usually stored in one of the following types of systems:

Databases: The primary goal of a database is to store data. It keeps records organized
and allows for quick lookups. However, it is not specifically designed for analysis.
Data Warehouses: A data warehouse is specifically designed for reading and
analyzing data. It combines data from multiple sources and organizes it so analysts can
easily query and explore it.
Data Marts: A data mart is a smaller subset of a data warehouse, focused on a specific
department or business function, such as sales, marketing, or finance.
Data Lakes: A data lake stores raw, unprocessed data in its natural format. It can hold
structured data like tables and unstructured data like images, emails, and documents.
Finding data inside a large organization can sometimes feel like detective work. As
companies grow, data gets spread across many systems. People leave the company or
change roles, and knowledge about where certain data lives can get lost.
Documentation is not always kept up to date. You may need to talk to many different
people across departments to track down the data you need.

Dealing with Data Access and Permissions

Even when you know where the data is, getting access to it can be difficult. Companies
protect their data carefully, and for good reason. Customer data, financial records, and
sensitive business information must be handled with care.

Many organizations have strict data access policies, sometimes called 'Chinese walls.'
These are rules that prevent people from accessing data they do not need for their job.
In most countries, there are also laws and regulations that govern how personal data
can be used. Getting permission to access the data you need may take time and may
involve going through official approval processes.

Looking Outside Your Organization

If the data you need does not exist internally, do not be afraid to look outside. There are
several options for finding external data:

• Commercial data providers such as Nielsen and GFK collect and sell high-quality
market research data, especially in the retail industry.
• Social media platforms like Twitter, LinkedIn, and Facebook provide APIs that
allow developers to access publicly shared data.
• Government and public organizations are increasingly sharing their data for free.
This open data often covers topics like population statistics, public health,
transportation, and economic indicators.

Some well-known open data sources include [Link] (US Government data), the
European Commission's open data portal, the World Bank data portal, and the US Food
and Drug Administration's database. These sources can provide high-quality, freely
available data that can complement your internal data or be used for training your data
science skills independently.

Performing Early Data Quality Checks

As soon as you retrieve data, you should do a quick quality check. This does not need
to be exhaustive at this stage, but it should help you catch obvious problems before you
invest too much time working with the data.

During data retrieval, check for the following:

• Does the data match what was described to you? Make sure you received the
right data set.
• Are the data types correct? For example, dates should be stored as date fields,
not text.
• Are there obvious missing values or completely empty columns?
• Does the number of records seem reasonable? Too few or too many records
may indicate a problem.

Early quality checks save enormous amounts of time later. In real-world data science
projects, data cleaning and correction can consume up to 80 percent of total project
time. Catching problems early means less rework down the line.
Step 3: Data Preparation

Data preparation is often called the most time-consuming step in data science. Real-
world data is rarely clean and ready to use. It comes with errors, inconsistencies,
missing values, and various other problems that need to be fixed before any meaningful
analysis can be done.

Think of data preparation like cooking. Before you can cook a great meal, you need to
wash the vegetables, chop them properly, measure the right amounts, and organize all
your ingredients. Only then can you begin actually cooking. Similarly, in data science,
you must prepare your data before you can start analyzing it or building models.

The Garbage In, Garbage Out Principle

In data science, there is a famous saying: 'Garbage in equals garbage out.' This means that
if you feed bad quality data into your models, you will get bad quality results — no matter
how sophisticated your model is. Data preparation is what separates good results from bad
ones.

Data Cleansing: Fixing Errors and Problems

The first part of data preparation is data cleansing. This involves identifying and
correcting all the errors and inconsistencies in your data. There are several common
types of errors you will encounter:

Data Entry Errors

These are mistakes made when data was originally entered, usually by humans typing
information manually. For example, a customer might be listed as living in 'USQ' instead
of 'USA,' or a product code might have a typo. These need to be identified and
corrected.

Physically Impossible Values

Sometimes data contains values that are logically impossible. For example, a person's
age recorded as 200 years, or a temperature reading of -500 degrees Celsius. These
values need to be removed or replaced.
Missing Values
Missing data is extremely common. A customer record might be missing an email
address or a phone number. You need to decide how to handle these gaps — either by
removing those records, filling in estimated values based on other data, or leaving them
as-is depending on the context.

Outliers
Outliers are values that are far outside the normal range. For example, if most
customers spend between 10 and 200 dollars per visit, but one customer has a
recorded spend of one million dollars, that might be an error or a very unusual case.
You need to investigate and decide how to handle outliers.

Formatting Inconsistencies
The same information might be recorded in different formats across different data
sources. For example, dates might be written as '01/03/2024' in one system and 'March
1, 2024' in another. Country names might be written as 'UK,' 'United Kingdom,' and
'Great Britain' in different records. All of these need to be standardized to a single
consistent format.

Data Transformation: Reshaping Your Data

After cleaning, the next step is data transformation. This involves changing the structure
or format of your data to make it more suitable for analysis and modeling. Common
transformation tasks include:

• Aggregating data — summarizing detailed records into higher-level summaries.

For example, combining daily sales figures into monthly totals.
• Extrapolating data — estimating missing values based on trends in the existing
data.
• Creating derived measures — calculating new variables from existing ones. For
example, calculating a customer's total annual spending from their monthly
transaction records.
• Creating dummy variables — converting categorical variables (like country
names or product categories) into numerical format so they can be used in
mathematical models.
• Reducing the number of variables — removing irrelevant or redundant columns
that will not add value to the analysis.

Combining Data from Multiple Sources

Often, the data you need is spread across several different sources. For example,
customer information might be in one database, purchase history in another, and
demographic data from an external source. Combining all of this into a single, unified
data set is a crucial part of data preparation.

This process involves merging or joining data sets, which can be complex if the different
sources use different identifiers for the same entity. For example, one system might
identify a customer by their email address, while another uses a customer ID number.
Linking records correctly across systems requires careful attention to detail.

By the end of data preparation, you should have a clean, organized, unified data set
that is ready to be explored and analyzed. This step is the unsung hero of data science
— the better you do here, the more reliable and accurate your results will be.
Step 4: Data Exploration

With clean, prepared data in hand, the next step is to explore it. Data exploration — also
called Exploratory Data Analysis, or EDA — is the process of getting to know your data
deeply. You look for patterns, trends, relationships, and anomalies that will inform how
you build your models later.

Data exploration is a bit like being a detective. You do not know exactly what you are
looking for yet. You are gathering clues, forming hypotheses, and building a deep
understanding of what the data is telling you. This step is partly analytical and partly
creative.

Why Exploration Matters

Skipping data exploration is one of the biggest mistakes a data scientist can make. Without
it, you might build a model based on incorrect assumptions, miss important patterns, or
overlook key variables that would dramatically improve your results. Exploration ensures
your modeling decisions are grounded in a real understanding of the data.

Visual Exploration Techniques

The most intuitive way to explore data is through visualization. Seeing data in graphical
form reveals patterns that would be impossible to spot by looking at numbers in a table.
Common visualization techniques include:

Simple Graphs
Bar charts, pie charts, line graphs, and histograms are basic but powerful tools. A
histogram, for example, shows how a variable is distributed — whether it is spread
evenly, skewed to one side, or concentrated around a particular value. A line graph can
reveal trends over time, such as whether sales are increasing or decreasing from month
to month.

Scatter Plots
Scatter plots show the relationship between two variables. If you plot customer age on
one axis and spending amount on the other, the scatter plot will reveal whether older
customers tend to spend more or less than younger ones. This kind of visual correlation
is extremely valuable when selecting which variables to include in your model.
Combined and Interactive Graphs
More advanced visualizations combine multiple data dimensions into a single view.
Interactive tools allow you to zoom in on specific subsets of data, filter by certain
criteria, and examine relationships from multiple angles. These are particularly useful
when working with large, complex data sets.

Link and Brush

This is a technique where selecting data points in one chart automatically highlights the
same points in other linked charts. It helps you see how the same records appear
across different variables, which can reveal complex multi-dimensional patterns.

Non-Graphical Exploration Techniques

In addition to visuals, data exploration also involves statistical analysis. This includes:

• Calculating summary statistics — mean, median, mode, minimum, maximum,

and standard deviation for each variable.
• Examining correlations — measuring how strongly two variables are related to
each other.
• Identifying distributions — understanding whether a variable follows a normal
bell-curve distribution or has a different shape.
• Detecting outliers — using statistical methods like the interquartile range to
identify values that are unusually far from the average.

What to Look For During Exploration

During data exploration, you are particularly looking for the following insights:

• Which variables are most related to the outcome you want to predict?
• Are there groups or clusters within the data that behave differently?
• Are there any surprising or unexpected patterns that need further investigation?
• Are there variables that are highly correlated with each other? (These might
cause problems in modeling and may need special handling.)
• Are there any remaining data quality issues that were not caught during data
preparation?
The insights you gather during exploration directly inform the decisions you make in the
next step — model building. Think of exploration as doing your homework before the
exam. The more thoroughly you understand your data, the better your models will be.
Step 5: Data Modeling

Data modeling is often considered the most exciting part of the data science process.
This is where you actually build the mathematical tools — called models — that can
make predictions, find patterns, or answer the questions defined in your research goal.
It is the step that most people think of when they imagine 'data science.'

However, without the solid foundation built in the previous four steps, even the most
sophisticated models will produce unreliable results. A model is only as good as the
data it is trained on and the understanding behind it.

Simple Models Often Win

Research has consistently shown that a combination of several simple models often
outperforms a single complicated model. This is an important lesson for aspiring data
scientists: do not always reach for the most complex technique. Start simple and increase
complexity only when necessary.

Model and Variable Selection

The first decision in data modeling is choosing which type of model to use. Different
types of problems call for different types of models:

Regression Models: Used when you want to predict a continuous numerical value. For
example, predicting the price of a house based on its size and location, or forecasting
next month's sales revenue.
Classification Models: Used when you want to predict which category something
belongs to. For example, predicting whether an email is spam or not spam, or whether a
loan applicant will repay or default.
Clustering Models: Used when you want to find natural groups within your data without
predefined categories. For example, segmenting customers into groups based on their
purchasing behavior.
Time Series Models: Used when you want to predict future values based on patterns in
historical data over time. For example, forecasting stock prices or electricity demand.

Alongside choosing the model type, you also need to decide which variables (also
called features) to include in the model. More variables are not always better. Including
irrelevant variables can actually hurt model performance by introducing noise. Variable
selection is both a science and an art.

Training the Model

Once you have chosen a model type and selected your variables, the next step is to
train the model. Training means showing the model a large set of examples — the
training data — and allowing it to learn the patterns in that data.

During training, the model adjusts its internal parameters to minimize the difference
between its predictions and the actual known outcomes. This process is repeated many
thousands of times until the model learns the underlying patterns as accurately as
possible.

It is important to note that you should never train and test your model on the same data.
If you do, the model will appear to perform brilliantly on the training data but will fail on
new, unseen data. This problem is called overfitting. To prevent it, data scientists
typically split their data into three sets: a training set, a validation set, and a test set.

Model Evaluation and Comparison

After training, you must evaluate how well the model performs. Different types of models
use different performance metrics:

• For regression models, common metrics include Mean Absolute Error (MAE),
Root Mean Square Error (RMSE), and R-squared, which measure how close the
model's predictions are to the actual values.
• For classification models, metrics include Accuracy, Precision, Recall, F1-Score,
and the Area Under the Curve (AUC), which measure how correctly the model
categorizes items.
• For clustering models, metrics such as the Silhouette Score measure how well-
defined the clusters are.

You should always try and compare multiple models before deciding on a final one. It is
rare that the first model you build is the best one. Trying different algorithms, different
combinations of variables, and different parameter settings often leads to significant
improvements.

Hyperparameter Tuning
Most machine learning models have settings called hyperparameters that control how
the model learns. These are not learned from the data; they are set by the data scientist
before training begins. Examples include the depth of a decision tree, the number of
trees in a random forest, or the learning rate in a neural network.

Finding the optimal hyperparameter settings for your model is called hyperparameter
tuning. This process can significantly improve model performance. Common techniques
include grid search, where you systematically try every combination of settings, and
random search, where you sample random combinations more efficiently.
Step 6: Presentation and Automation

You have done the hard work. You have defined a clear goal, gathered and cleaned
your data, explored it thoroughly, and built a well-performing model. Now comes the
final but crucial step: sharing your results with the world and, where needed, automating
the process so it runs continuously without manual effort.

Many data scientists underestimate the importance of this step. You might have
discovered truly valuable insights, but if you cannot communicate them clearly and
convincingly, they will have no impact on the business. The ability to present findings in
a simple, compelling way is one of the most valuable skills a data scientist can develop.

The Art of Communication

A data science project is only truly successful when it changes the way decisions are made
in an organization. No matter how accurate your model is, if the people who need to act on it
do not understand or trust it, nothing will change. This is why presentation skills are just as
important as technical skills.

Presenting Your Results

When presenting results to a non-technical audience, the key is to translate complex
technical findings into simple, actionable business language. Here are the principles of
good data science presentation:

8. Start with the conclusion — tell your audience what you found before explaining
how you found it. Business people are time-pressed and want to know the
bottom line immediately.
9. Use visuals generously — charts, graphs, and infographics are much easier to
understand than tables of numbers or pages of text. A single well-designed chart
can convey what would otherwise require paragraphs of explanation.
10. Connect findings to business impact — always explain what your findings mean
in terms of money saved, revenue gained, customers retained, or risks avoided.
This is what your audience truly cares about.
11. Keep it simple — avoid jargon. Do not say 'the model achieved an AUC of 0.87.'
Instead, say 'our model correctly identifies 87 out of every 100 customers who
are at risk of leaving.'
12. Tell a story — structure your presentation as a narrative with a beginning (the
problem), middle (the analysis), and end (the solution and recommendation).
Convincing Stakeholders
One important aspect of presentation is convincing the business or organization that
your findings are trustworthy and that acting on them will produce the expected results.
This requires more than just showing good numbers.

You may need to explain your methodology in a way that builds confidence without
overwhelming your audience with technical details. You may also need to address
concerns or objections. Perhaps a manager does not trust machine learning and prefers
human judgment. Or perhaps there are budget constraints that limit how the model can
be implemented. Being prepared for these challenges and addressing them calmly and
confidently is part of your role as a data scientist.

Automating the Analysis

For many data science projects, the analysis does not happen just once. It needs to be
repeated regularly as new data arrives. For example, a model that predicts which
customers are likely to cancel their subscription needs to run every week so the sales
team always has an up-to-date list to work from.

Automating this process means writing code or setting up systems that can perform the
entire data science pipeline — from data collection to model prediction — automatically,
on a schedule, without human intervention. This is a critical skill in modern data science.

Automation typically involves:

• Setting up data pipelines that automatically pull fresh data from source systems
on a regular schedule.
• Running data cleaning and preparation scripts automatically when new data
arrives.
• Executing the trained model to generate new predictions and store them in a
database.
• Generating automated reports or dashboard updates so stakeholders always
have access to current insights.
• Monitoring the model's performance over time and sending alerts if performance
drops below an acceptable level.

Well-automated data science processes can save businesses enormous amounts of

time and money. They ensure that insights are always fresh and available, without
requiring a data scientist to manually run the analysis every week.
Important Concept: Don't Be a Slave to the Process

Throughout this document, we have described the six-step data science process as if it
always progresses in a neat, linear fashion from Step 1 to Step 6. In reality, this is never
the case. Real data science projects are messy, iterative, and unpredictable. You will
frequently find yourself jumping back to earlier steps as you discover new information.

A Key Warning
The six-step process is a guide, not a rigid set of rules. Your process will be shaped by your
preferences, your company's culture, the nature of the project, and the constraints you face.
Treat it as a map, not a script.

The Reality of Iteration

Here is a typical example of how iteration happens in a real project. You complete data
preparation and move on to data exploration. While creating graphs, you notice a clear
difference in behavior between two groups of customers — perhaps men and women.
But then you realize that your data set does not actually include a column that indicates
gender. You need to go back to Step 2 and retrieve additional data.

But getting that additional data is not as simple as just downloading it. You need to go
through your organization's data approval process, which may require writing a new
mini project charter and getting sign-off from management. This entire detour might take
days or weeks.

When to Use a Flexible Approach

The agile methodology — a flexible, iterative approach originally developed for software
development — is increasingly being adopted by data science teams. Rather than
completing every step fully before moving on, agile data science involves working in
short cycles, producing working results quickly and refining them over time.

The agile approach works particularly well when the requirements are unclear at the
start, when the business needs quick results before a full solution is ready, or when the
project involves a lot of experimentation and uncertainty. However, many large
organizations still prefer a more structured, sequential approach because it is easier to
manage, budget, and audit.
Working in Prototype Mode
One of the key advantages of the structured process is that it encourages working in
prototype mode. When building a prototype, you focus on finding the best model
quickly, without worrying about things like code optimization, documentation standards,
or production-readiness. This allows you to move fast and experiment freely.

Once you have found a model that works well in prototype form, you can then invest
time in cleaning up the code, optimizing performance, and preparing it for deployment in
a production environment. This two-phase approach — rapid prototyping followed by
production hardening — is a practical and efficient way to deliver data science projects.
The Importance of Teamwork in Data Science

One of the most important benefits of following a structured data science process is that
it enables team collaboration. Data science is not a one-person job. It requires a diverse
set of skills, and it is virtually impossible for any single person to be an expert in all of
them.

Consider everything that a data science project requires: deep knowledge of database
management and SQL, expertise in statistical methods and machine learning,
programming skills in Python or R, domain knowledge about the specific industry,
communication and presentation skills, project management experience, and an
understanding of data privacy and legal regulations. No one person can be a master of
all these areas.

Specialization is Strength
By dividing a project into clearly defined steps, different specialists can contribute their
expertise where it matters most. Database engineers handle data retrieval, statisticians
handle exploration and modeling, software engineers handle deployment and automation,
and business analysts handle definition and presentation. The sum is far greater than its
parts.

Roles in a Data Science Team

A modern data science team typically includes several different roles, each contributing
to different steps of the process:

Data Engineer: Responsible for building and maintaining the infrastructure that stores
and processes data. They handle data pipelines, databases, and cloud systems. They
are most active in Steps 2 and 3.
Data Analyst: Focuses on exploring and interpreting data to find patterns and insights.
They are most active in Steps 3 and 4, producing reports and visualizations that inform
decision-making.
Data Scientist: Builds and evaluates predictive models. They combine statistical
knowledge with machine learning expertise and are most active in Steps 4 and 5.
Machine Learning Engineer: Takes the models built by data scientists and deploys
them into production environments, making them available at scale. They are most
active in Steps 5 and 6.
Business Analyst: Bridges the gap between technical teams and business
stakeholders. They help define the research goal, interpret findings in business terms,
and communicate results effectively. They are most active in Steps 1 and 6.

When each team member understands their role within the broader process and
communicates effectively with the rest of the team, the entire project flows more
smoothly and the results are more impactful.
Summary: The Six Steps at a Glance

Here is a concise summary of all six steps of the data science process, what they
involve, and what they produce:

Step Name What You Do What You Produce

Step 1 Setting Research Define what, why, Project Charter
Goal and how of the
project
Step 2 Retrieving Data Find, access, and Raw Data Set
collect relevant data
Step 3 Data Preparation Clean, transform, Clean, Ready Data
and combine data
Step 4 Data Exploration Visualize and Insights &
analyze patterns in Hypotheses
the data
Step 5 Data Modeling Build, train, and Trained Model
evaluate predictive
models
Step 6 Presentation & Share findings and Reports &
Automation automate the Automated System
pipeline

The Flow Diagram in Simple Terms

Think of the six steps as a cycle, not just a straight line:

13. Ask the right question and plan your project.

14. Gather the data you need to answer that question.
15. Clean and organize the data so it is ready to use.
16. Explore the data to understand what it is telling you.
17. Build a model to make predictions or find patterns.
18. Share your results and set up the system to run automatically.
And at any point, be ready to loop back to an earlier step if new information or
challenges arise. The most successful data scientists are not the ones who follow the
process most rigidly, but the ones who understand the purpose behind each step and
apply good judgment at every stage.
Conclusion

The data science process is a powerful framework that helps data scientists and their
teams work in an organized, efficient, and effective way. By following the six steps —
setting the research goal, retrieving data, preparing data, exploring data, building
models, and presenting results — you dramatically increase your chances of producing
valuable, reliable, and actionable insights.

Each step serves a specific and important purpose. The first step ensures everyone is
aligned on what the project is trying to achieve. The second step ensures you have the
right data. The third step ensures the data is clean and reliable. The fourth step ensures
you deeply understand the data before committing to a model. The fifth step is where
the actual prediction or discovery happens. The sixth step ensures the insights reach
the people who can act on them.

Perhaps most importantly, remember that this process is not a rigid set of rules. It is a
guide. Real projects require iteration, flexibility, and good judgment. You will go back
and forth between steps. You will encounter unexpected challenges and make
discoveries that change your direction. Embrace this uncertainty — it is part of what
makes data science both challenging and exciting.

The best data scientists combine technical expertise with strong communication skills,
business understanding, curiosity, and the ability to work effectively in a team. The six-
step process provides the structure within which all of these qualities can flourish.

Final Thought
Data science is ultimately about improving the world through better decisions. Every clean
data set, every well-built model, and every clearly communicated insight has the potential to
help a business serve its customers better, help a hospital treat its patients more effectively,
or help a government allocate its resources more wisely. The process you have learned in
this document is the foundation upon which all of that good work is built.

End of Document — PVP Siddhartha Institute of Technology, Dept. of IT