IBM Data Science Professional Certificate study plan

Define the data science lifecycle stages (problem framing, data acquisition, wrangling, EDA, modeling, evaluation, deployment, communication) and map each stage to the tool that performs it in this curriculum.

Write and run Python scripts that use lists, dictionaries, tuples, sets, loops, conditionals, and functions with parameters and return values, then refactor one script to remove duplicated logic.

Distinguish supervised from unsupervised learning and classification from regression, and match five realistic business questions to the correct problem type and evaluation metric.

Set up a local Python environment (Anaconda or venv) and verify it by installing pandas, numpy, and matplotlib and importing them in a script.

Manipulate NumPy arrays: create them, reshape, slice, broadcast, and apply vectorized arithmetic instead of Python loops.

Load a CSV into a pandas DataFrame, inspect it with head, info, describe, and dtypes, and document its shape, column types, and missing-value counts.

Select, filter, and sort data with loc, iloc, boolean masks, and query, and derive new columns from existing ones.

Handle missing data deliberately by choosing between dropping, mean/median imputation, and forward fill, and justify each choice in writing.

Group and aggregate data with groupby, agg, and pivot_table to answer three specific questions about the dataset.

Merge and concatenate multiple DataFrames using inner, left, and outer joins, and explain in writing how row counts change after each join.

Write SQL SELECT statements with WHERE, ORDER BY, GROUP BY, HAVING, and aggregate functions against a sample relational database.

Connect to a SQL database from Python (sqlite3 or SQLAlchemy), pull a result set into a DataFrame, and compare the same aggregation done in SQL versus pandas.

Perform exploratory data analysis that computes summary statistics, correlations, and value counts, and state three hypotheses the data can test.

Implement simple and multiple linear regression with scikit-learn, split data into train and test sets, and report R-squared and RMSE.

Build and evaluate a classifier (logistic regression or decision tree) and interpret a confusion matrix, accuracy, precision, and recall.

Explain overfitting and underfitting and demonstrate one regularization or cross-validation technique to mitigate them.

Tune hyperparameters with GridSearchCV and report how the best parameters changed validation performance.

Create line, bar, scatter, and histogram plots with matplotlib and pandas, and label axes, titles, and legends correctly.

Build a multi-panel figure with seaborn that reveals a relationship between at least three variables without misleading scales or truncated axes.

Critique two charts for misleading design choices and rewrite each caption to state the actual finding.

Build an interactive dashboard in a notebook or a dashboarding tool with at least two filters and three linked charts.

Design a colorblind-safe palette and apply consistent styling across all charts in the dashboard.

Write three analytical SQL queries that feed dashboard metrics, including one with a subquery or window function.

Present the dashboard's key insight in three bullet points aimed at a non-technical stakeholder.

Define a capstone question, identify a public dataset, and document scope, success criteria, and known limitations before analysis.

Execute the full pipeline on the capstone data: acquisition, wrangling, EDA, modeling, and evaluation, committing work to GitHub in logical increments.

Use Git commands (clone, add, commit, push, branch) and write a README that lets a stranger reproduce your environment and results.

Run the capstone notebook end to end from a clean kernel and fix any hidden state or ordering dependencies.

Produce a final capstone report that states the problem, method, results, limitations, and recommended next steps in plain language.

Deliver a five-minute recorded or live presentation with slides that lead with the finding rather than the process.

Answer three likely stakeholder objections about your model's assumptions, data quality, or generalizability with prepared responses.

Complete a timed 40-question mixed-domain practice set and log every missed concept into a review list for final revision.