Start here · your first 30 minutes

Welcome to ISOM 835

This course teaches one skill, thirteen times: take a business question about the future, turn it into a model in Python, evaluate it honestly, and turn the prediction into a decision someone would act on. This page is everything you need before the first class.

Mondays · 5:00 PM – 7:30 PM · Prof. Hasan Arslan, PhD · harslan@suffolk.edu

01 // SETUP

Four things, eleven minutes

Nothing to install. Do these once, before Session 1 if you can, before Session 2 at the latest.

Google Colab
2 min

Every session notebook opens in Colab from the session page with one click. Sign in with any Google account; nothing to install. Tip: File → Save a copy in Drive the first time you open a notebook, so your edits persist.

Open Colab →
Kaggle account
3 min

You need one for the Midterm Model Competition (launches Oct 19) and it is the best free source of project datasets. Use your suffolk.edu email so we can find your team.

Create an account →
The course repository
1 min

All notebooks, datasets, and this site live in one public GitHub repo. You never need to clone it — but star it, and read a notebook there when Colab is slow.

Open the repo →
An AI assistant (optional, encouraged)
5 min

Colab has a built-in Gemini assistant; Claude and ChatGPT work in a second tab. Use them the way professionals do — to draft, then verify the split, the leakage, the metric, and the baseline yourself. Disclose what you used.

How we teach this (Session 12) →
02 // RHYTHM

How every week works

The same shape, thirteen times. About three hours outside class per week; more in project weeks.

Before class
30–45 min
  • Read the session page: objectives and agenda (5 min).
  • Do the reading — usually one ISLP or Géron chapter section, always free.
  • Open the notebook in Colab and run the first three cells so the data is loaded when you arrive.
In class
150 min
  • A business question, a dataset, and a model you build alongside the instructor.
  • The Explorer for the week, on your phone or laptop, when a concept needs a picture.
  • A 10-minute break at 6:05. Coffee is downstairs.
After class
60–90 min
  • Finish the three "Your turn" exercises at the end of the notebook.
  • Do the Practice questions for that domain — six questions, misconception-tagged.
  • Homework is due on Canvas at 11:59 PM on its due date — always a Monday, so you can ask questions in class first. One thing per week: see every date on the syllabus, or add the whole semester to your calendar from there.
Your progress · saved on this device
Where you are
Nothing recorded yet — start with Python, Guided or a Practice question.
Python, Guided0 / 28
Practice bank0 / 36 correct
03 // THE SITE

Seven places you'll live

Python Track
Python, Guided: seven lessons you read, run, and get graded on in the browser. Do it before Session 2 if pandas is new to you.
Sessions
One page per Monday: objectives, agenda, the notebook, readings, homework.
Explorers
Draggable versions of the big ideas. If you can predict what the picture does before you drag, you understand it.
Practice
36 exam-style questions. Every wrong answer names the misconception behind it.
Cheat Sheet
The whole course on one page: templates, formulas, checklists. Print it.
Datasets
Every dataset we use, with a copy-paste loader that needs no login.
Project
Your Prediction Story: the brief, the rubric, the deadlines, the FAQ.
04 // FIVE HABITS

What separates a model that ships from one that merely runs

These come up in every session and every rubric. Learn them early and the rest of the course is details.

1
Frame it in four lines
Unit, target, horizon, decision. Write them at the top of every notebook and every memo. Session 1.
2
Beat the baseline first
"Predict the majority class" or "same hour last week" is the score to beat. If you cannot name the baseline, you cannot claim a win. Session 1, 11.
3
Split before you fit anything
Scalers, imputers, encoders learn from data. Put them in a Pipeline fit on the training fold. Session 3.
4
Ask what a mistake costs
A model gives a score; the business needs a decision. The threshold comes from the cost matrix, not from 0.5. Session 5.
5
Earn trust three ways
Why (SHAP), how sure (calibration, conformal), where it fails (sliced error rates). Session 9.
05 // REFRESHERS

If Python or pandas feel rusty

Prof. Arslan's earlier ISOM 730 notebooks, hosted in this repo so they open in Colab with one click. Work through the first two before Session 2.

pandas & exploratory data analysis
Series, DataFrames, joins, groupby — the pandas you need for Session 2.
Open in Colab →
matplotlib & seaborn
Line, scatter, histogram, pair plots — the charts that decide which features to engineer.
Open in Colab →
Introducing scikit-learn
The estimator API: choose, configure, fit, predict. The slow version of Session 1's twelve lines.
Open in Colab →
Model validation & selection
Holdout, cross-validation, bias–variance, grid search — the ideas behind Sessions 1, 6, and 9.
Open in Colab →
Linear regression, ridge & lasso
Basis functions and regularization on a small example. Pairs with Session 6.
Open in Colab →
Decision trees & random forests
A tree on blobs, the overfitting picture, bagging. Pairs with Session 7.
Open in Colab →
Müller & Guido, Ch. 2 — Supervised learning (official book notebook)
Every classifier with its decision boundary drawn, by a scikit-learn core developer. Older scikit-learn, timeless ideas.
Open in Colab →
Müller & Guido, Ch. 5 — Evaluation & improvement (official book notebook)
Cross-validation, grid search, and the metrics chapter. Pairs with Sessions 5, 6, and 9.
Open in Colab →
06 // YOUR FIRST THREE MONDAYS

Where to begin

Session 1 · Mon Sep 14
Foundations
Open the session page →
Session 2 · Mon Sep 21
Data & Features
Open the session page →
Session 3 · Mon Sep 28
Data & Features
Open the session page →

One promise

By December you will be able to take any business question about the future, frame it, build a model that answers it, evaluate it honestly, and hand a decision-maker a number they can act on — and explain every line of it without an AI standing next to you.