Stop 09 · Advanced Analytics & Predictive BasicsTopics · 6/8

Intro to Machine Learning for Analysts

In short

Machine learning sounds like magic, but for analysts it mostly means letting a model learn patterns from examples instead of writing every rule yourself. Knowing the basics helps you use it wisely.

A conceptual overview of what ML models are (pattern-matching functions trained on data) and where analysts fit in — usually preparing data and validating outputs, not building models from scratch.

The jaadu shows a ripe and an unripe mango to a small curious robot.

Think of it like this

Teaching a robot about mangoes

You could write rules: yellow, soft, sweet smell means ripe. Or show a little robot a hundred mangoes labelled ripe or not and let it work out the pattern. That second way is machine learning. It works well with lots of good examples and falls over with bad ones.

Watch it happen

Rules you write vs patterns it learns

Hand-written rules

Easy to explain and audit
Break when patterns get complex
Need updating by hand

Machine learning

Learns patterns from labelled examples
Handles many inputs at once
Needs good data and testing

Start with simple rules or a simple model. Reach for ML when rules get messy and you have plenty of labelled examples.

The big ideas

What you need to know

Supervised learning

Examples come with answers, like churned or not. The model learns to predict the answer. Regression and classification live here.

Unsupervised learning

No answers given. The model finds structure, like clusters of similar customers.

Train, test, repeat

Fit on one part of the data, judge on another. Overfitting means great on training, poor on new data.

Explainability matters

Business users trust what they understand. A simple model you can explain often beats a black box.

See it in code

Walk through the code

python

1from sklearn.model_selection import train_test_split2from sklearn.tree import DecisionTreeClassifier3 4X = df[["weight_g", "softness", "colour_score"]]5y = df["ripe"]6X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.3, random_state=0)7 8for depth in [2, 5, None]:9    tree = DecisionTreeClassifier(max_depth=depth, random_state=0).fit(X_train, y_train)10    print(depth, "train:", round(tree.score(X_train, y_train), 2),11          "test:", round(tree.score(X_test, y_test), 2))

1/3Inputs, labels, and a 70/30 train/test split.

The same model at three depths. Watch training accuracy rise while test accuracy stalls or drops: that's overfitting.

Save yourself the pain

Common mistakes

Oops

Judging a model by its accuracy on the training data.

Do this instead

Always report performance on a held-out test set, ideally from a later time period.

Oops

Using ML because it sounds impressive.

Do this instead

Try a simple baseline first, like last month's value or a rule. ML must beat it to earn its place.

Your mission

See overfitting yourself

~30 min

Quick check

A model scores 99% on training data and 70% on test data. What's most likely happening?

Go deeper

1Read a high-level explainer distinguishing supervised, unsupervised, and reinforcement learning.
2Identify what a data scientist would need from you to build a model well.
3Learn to critically evaluate a model's output, not just accept it as ground truth.