Blog by group CONQ026 — Warm Up 02

Introduction

In this article, we'll walk through the complete picture: what Data Science is, what the work actually looks like day-to-day, what a standard project process looks like, and how the roles in the field truly differ from one another. No complex jargon, no dry theory, just the things that are genuinely useful.

1. What is Data Science - and where is it in the AI landscape?

Before understanding what Data Science is, let's clarify 4 concepts that most beginners confuse: AI, Machine Learning, Deep Learning, and Data Science.

Looking at the diagram below, you'll notice something interesting right away: Data Science does not sit inside AI, it's its own circle, with overlapping areas but completely independent.

• Artificial Intelligence (AI) - The largest circle

AI is the broadest field, with the goal of simulating human intelligence so that computers can think, learn, and make decisions on their own. Everything below it, ML and Deep Learning, falls within AI.

• Machine Learning (ML) - Inside AI

ML is a branch of AI. Instead of hard-coding every rule, ML allows computers to learn on their own from data. For example: instead of writing code that says "if an email contains the word 'promotion' then it's spam", ML learns to recognize spam on its own after seeing thousands of email samples.

• Deep Learning (DL) - Inside ML

DL is a branch of ML that uses multi-layered neural networks to tackle more complex problems like facial recognition, language translation, or image generation. This is the technology behind ChatGPT, Midjourney, and the generative AI models you see today.

• Data Science - An independent circle

Data Science does not sit inside AI. Data Science is a field that combines mathematics, statistics, programming, and domain knowledge to collect, process, and analyze data, with the goal of uncovering valuable insights and making better decisions.

DS overlaps with ML and DL, because a Data Scientist can use ML/DL as analytical tools. But a skilled Data Scientist can create enormous value without ever touching Deep Learning, relying solely on statistics, SQL, and the ability to tell stories with data.

The table below summarizes the differences between all four concepts:

2. What does the actual work in Data Science look like?

Because Data Science is inherently interdisciplinary, a Data Scientist's work spans a wide range of tasks - from data collection and data processing to analysis and visualization. These tasks form a pipeline that transforms raw data into meaningful insights.

2.1. Data Collection

The golden rule of the field: "Garbage In, Garbage Out." Poor input data means unreliable results, no matter how sophisticated the algorithm.

Collecting data well means knowing clearly what you need, where to get it, and documenting every step carefully. Data sources can include internal databases, web scraping, APIs, user surveys, or publicly available datasets.

2.2. Data Cleaning

This is the part that never gets mentioned in Data Science marketing materials, yet it's the most time-consuming step - typically accounting for 60% to 80% of total project time.
Real-world data is always messy. Missing values, duplicate records, inconsistent formatting ("HCM", "Ho Chi Minh", "HCMC", three ways to write the same thing), and outliers that skew results. All of it must be detected and handled before any analysis can begin.

2.3. Exploratory Data Analysis & Data Visualization

Once the data is clean, it's time to start "reading" it. Exploratory Data Analysis (EDA) is the process of understanding the characteristics of your data, how it's distributed, how variables relate to each other, and whether any anomalies remain.

In EDA, you'll be looking to answer questions like:

• How is the data distributed? Is it skewed or balanced?

• Are there correlations between variables? Which ones influence the target variable?

• Are there any outliers that slipped through the Cleaning step?

• Are there any interesting patterns or trends hidden in the data?

Visualization is the primary tool of EDA. Each chart type lets you see the data from a different angle, as illustrated in the figure below:

2.4. Building Predictive Models (Machine Learning)

This is the step most people think of first when they hear "Data Science", but in reality, it takes up only a small fraction of total project time. Commonly used ML algorithms include:

🔑 Key takeaway
These four steps are not a straight line — they form a loop. Poor model results may send you back to clean more data, or collect new data entirely. This is the real nature of the work.

3. The standard Data Science project process: CRISP-DM

Knowing the techniques isn't enough, you also need to know how to organize a project so it actually solves the business problem. That's why CRISP-DM exists.

CRISP-DM (Cross-Industry Standard Process for Data Mining) is the most widely used industry-standard process for executing Data Science and Machine Learning projects. It was developed in 1996 by a consortium including Daimler, SPSS, and NCR.

It consists of 6 sequential stages: understanding the business problem, understanding and preparing the data, building the model, evaluating results, and deploying the product. The process is iterative - if results don't meet expectations, you loop back to earlier stages to improve.

4. Roles in the Data field - Which one suits you?

In a data or AI project, no one does everything alone. Each role is a link in the chain from raw data to business decision. Here's the full picture:

4.1. Overview of the 5 core roles

4.2. The 3 roles you should understand first

4.2.1. Data Analyst (DA)

DA is the bridge between data and decision-makers. They use SQL to extract data, Python or Excel to analyze it, and Power BI or Tableau to turn dry numbers into stories that leadership can understand and act on.

• Real-world example: A fashion chain notices a sudden drop in revenue in March. The DA investigates, queries the data, and discovers: revenue fell mainly in winter clothing because warm weather arrived earlier than usual that year. They then recommend launching the summer collection ahead of schedule - a decision driven directly by data.

4.2.2. Data Scientist (DS)

DS goes further than DA by building models capable of predicting the future. They need a stronger foundation in mathematics and statistics, and typically work more deeply with machine learning algorithms.

If a DA answers "Why did revenue drop last month?", a DS answers "Which customers are most likely to churn in the next 3 months?"

• Real-world example: A bank wants to reduce lending risk. The DS builds a credit scoring model based on transaction history, age, and occupation of millions of past customers. When a new applicant submits a loan request, the model automatically calculates a risk probability and supports the bank's decision in just seconds.

4.2.3. Data Engineer (DE)

If DS and DA are the "users" of data, DE is the one who makes sure that data exists, is clean, and is ready to use. They build pipelines that automate the collection, processing, and storage of data at scale.

• Real-world example: A food delivery app has thousands of drivers moving constantly. The DE builds an automated pipeline to collect driver GPS coordinates every second, filter out bad data, and load it into the data warehouse - so the DS can immediately use it to calculate estimated delivery times for millions of orders per day.

Summary table of the 3 core roles:

5. Where should you start?

After reading this far, your question should be: "So which direction should I go?" Here's a quick suggestion based on personality and interests:

Practical advice:
No matter which direction you choose, SQL and basic Python are two non-negotiable foundational skills. Start here before worrying about anything else.

Wrapping Up

Data Science is not a single path. It's an ecosystem with many roles, many directions, and many ways to create value from data.

This article is the big-picture overview, but it's only the starting point. Once you've identified the direction you want to go, the next step is to go deeper into each role, each tool, and each specific step in the learning roadmap.