Data scientists turn messy real-world data into useful evidence for decisions. Their work connects computer science, statistics, domain knowledge, and communication. A typical day is not just building machine learning models, because much of the job is understanding data, cleaning it, checking assumptions, and explaining results.
This matters because organizations use data science to improve products, reduce risk, detect patterns, and choose better actions.
A common workflow starts with collecting data from databases, apps, sensors, surveys, or web logs, then using SQL and Python to prepare it for analysis. Data scientists explore patterns with summaries and visualizations, build models when prediction is useful, and test ideas with experiments such as A/B tests. They often create dashboards so teams can monitor key metrics over time.
The final step is communication, where technical findings become clear recommendations for people who may not write code.
Key Facts
- Data science workflow: collect data, clean data, explore patterns, model or test, communicate results, monitor outcomes.
- Cleaning and preparing data can take about 80% of a data scientist's time because real data often has missing values, duplicates, errors, and inconsistent formats.
- Mean formula: mean = sum of values / number of values.
- A/B test lift formula: lift = (conversion rate B - conversion rate A) / conversion rate A.
- Model error example: MAE = sum(|actual - predicted|) / n.
- Core tools often include SQL for querying data, Python for analysis and modeling, notebooks for experiments, and dashboards for reporting.
Vocabulary
- Data cleaning
- Data cleaning is the process of fixing or removing missing, duplicated, inconsistent, or incorrect data before analysis.
- Exploratory data analysis
- Exploratory data analysis is the use of summaries, charts, and comparisons to understand patterns and problems in a dataset.
- Machine learning model
- A machine learning model is a program that learns patterns from data to make predictions or classifications on new examples.
- A/B test
- An A/B test is an experiment that compares two versions of something to see which performs better using measured outcomes.
- Dashboard
- A dashboard is a visual display of key metrics and charts that helps people monitor data and make decisions.
Common Mistakes to Avoid
- Jumping straight to machine learning, because a model built on poorly understood or dirty data can produce misleading predictions.
- Ignoring missing values, because blanks are not always random and can bias averages, trends, and model results.
- Confusing correlation with causation, because two variables moving together does not prove that one caused the other.
- Reporting results without business context, because accurate calculations are less useful if they do not answer the decision that needs to be made.
Practice Questions
- 1 A dataset has 50,000 rows. During cleaning, a data scientist removes 1,250 duplicate rows and 750 rows with unusable missing values. How many rows remain, and what percent of the original dataset was removed?
- 2 In an A/B test, version A has 2,000 visitors and 160 purchases. Version B has 2,500 visitors and 225 purchases. Find the conversion rate for each version and the lift of B compared with A.
- 3 A dashboard shows that app usage and ice cream sales both rise during summer. Explain why this pattern alone is not enough to conclude that ice cream sales cause app usage to increase.