What is data science? A practical roadmap (CRISP-DM made simple)

Data analysis explains what happened and why. Data science goes a step further: it uses statistics and machine learning to predict what is likely to happen next and to recommend what to do about it.

The CRISP-DM process

CRISP-DM (Cross-Industry Standard Process for Data Mining) is a widely used, tool-neutral framework. Its six phases loop back on each other:

1. Business understanding: what decision will this help? What does success look like?
2. Data understanding: what data exists, how good is it, what patterns appear?
3. Data preparation: cleaning, combining and shaping data. This often takes most of the time.
4. Modelling: trying techniques such as regression, classification or clustering.
5. Evaluation: does the model actually meet the business goal, not just score well?
6. Deployment: putting results to work in a report, an app or a process, and monitoring them.

A first predictive model

from sklearn.model_selection import train_test_split
from sklearn.linear_model import LogisticRegression

X = customers[["months_active", "orders", "support_tickets"]]
y = customers["churned"]          # 1 = left, 0 = stayed
X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.2, random_state=42)

model = LogisticRegression().fit(X_train, y_train)
print("Accuracy:", model.score(X_test, y_test))

Accuracy alone can mislead: if only 5% of customers churn, a model that always says "stays" scores 95%. Check precision and recall too, and ask whether the result changes a decision.

A beginner roadmap

1. Spreadsheet analysis and charts (spreadsheet skills).
2. Python and pandas (Python learning path).
3. Basic statistics: averages, distributions, correlation vs causation.
4. SQL for getting data (databases and SQL).
5. scikit-learn for your first models, then one end-to-end project for your portfolio.

Further reading

• CRISP-DM explained (Udacity)
• scikit-learn: getting started