Course Syllabus
ISOM 835 merges predictive analytics with machine learning, using Python for practical, real-world applications. Students learn the full business-analytics lifecycle — framing a prediction problem, preparing data, engineering leak-proof features, building and tuning regression, classification, ensemble, unsupervised, and forecasting models, evaluating them honestly, and turning predictions into decisions with a quantified business impact. The course reflects the 2025–26 state of practice: gradient boosting and tabular foundation models, conformal prediction, explainability, fairness slicing, forecasting foundation models, LLM-derived features, and responsible deployment under the EU AI Act and US state law.
No class Mon Oct 12 (October Break). Session 13 and final presentations meet in the exam-week slot, Mon Dec 14. No written final exam — the project is the final.
By December you will be able to
Thirteen Mondays
Nine modules. Each row links to the full session page with the notebook, readings, videos, explorer, homework, and practice questions.
| # | Date | Session | Reading before class | Due this week (Mon 11:59 PM unless noted) |
|---|---|---|---|---|
| Foundations · The prediction lifecycle — and your first model, end to end. | ||||
| 01 | Mon Sep 14 | From Data to Decisions: Your First Predictive Model | ISLP, Chapter 2 | — |
| Data & Features · Wrangling, exploration, and leak-proof feature pipelines. | ||||
| 02 | Mon Sep 21 | Data Wrangling & EDA for Prediction | pandas User Guide | — |
| 03 | Mon Sep 28 | Feature Engineering & Leak-Proof Pipelines | scikit-learn User Guide | — |
| Classification & Decisions · Predicting yes/no — then turning probabilities into profitable decisions. | ||||
| 04 | Mon Oct 5 | Classification: Predicting Yes/No | ISLP, Chapter 4 | HW1 — EDA + baseline (Telco) · Lab 0 Python Foundations Check |
| 05 | Mon Oct 19 | From Scores to Decisions: Thresholds, Costs & Calibration Competition launches | Google ML Crash Course | HW2 — credit default classification |
| Regression · Predicting numbers: linear models, regularization, and what a residual is telling you. | ||||
| 06 | Mon Oct 26 | Regression: Predicting Numbers | ISLP, Chapter 3 | — |
| Trees & Ensembles · Random forests, gradient boosting, and the tabular frontier. | ||||
| 07 | Mon Nov 2 | Decision Trees & Random Forests | ISLP, Chapter 8 | Project Milestone 1 — problem + dataset · competition team registered on Kaggle |
| 08 | Mon Nov 9 | Gradient Boosting & the Tabular Frontier Competition reveal · memo due | explained.ai | Competition: leaderboard closes 4:30 PM, predictions on Canvas · memo tonight |
| Evaluation & Optimization · Tuning, explaining, and trusting a model. | ||||
| 09 | Mon Nov 16 | Trust, but Verify: Tuning, Explaining & Uncertainty | Angelopoulos & Bates | HW3 — threshold/cost + Ames regression |
| Unsupervised Learning · Segments, structure, and anomalies — learning without labels. | ||||
| 10 | Mon Nov 23 | Segments, Structure & Anomalies | ISLP, Chapter 12 | Project Milestone 2 — baseline + metric |
| Time & Text · Forecasting, neural networks, embeddings, and LLMs as features. | ||||
| 11 | Mon Nov 30 | Forecasting: Time Series With ML & Foundation Models | Hyndman & Athanasopoulos | Project Milestone 3 — tuned model + error analysis |
| 12 | Mon Dec 7 | Neural Networks, Embeddings & LLMs as Features | Grinsztajn et al. (2022) & LLMTabBench (2026) | Project Milestone 4 — dry-run slides by 5:00 PM |
| Finale · Ship it — and present your prediction story. | ||||
| 13 | Mon Dec 14 | Ship It — Deployment, Monitoring & Your Prediction Story Final presentations · Signal Award | Project notebook + memo + repo (5:00 PM) · presentations tonight | |
Every due date, once
One thing due per week, always a Monday at 11:59 PM unless noted; nothing the day after class, nothing over Thanksgiving, nothing in the two days before the final slot. Each row opens the Canvas page.
How the grade is built
| A | 93–100 |
| A− | 90–92 |
| B+ | 87–89 |
| B | 83–86 |
| B− | 80–82 |
| C+ | 77–79 |
| C | 73–76 |
| F | below 73 |
What each piece is, and how it is graded
Participation & in-class labs
15%Every session has a live notebook. Credit for being present, running the lab, and completing the "Your turn" exercises by the following Sunday. Two lowest weeks dropped. Lab 0, the Python Foundations Check (due Mon Oct 5), is a completion credit here, not one of the three graded homework notebooks.
- Present and engaged (attendance, questions, explorer use)
- "Your turn" cells attempted in ≥ 10 of 12 lab sessions
Homework (3 notebooks, individual)
25%Three Colab notebooks on real data, each due at 11:59 PM on the due date, on Canvas. Graded on the same four things every time. The hotel-bookings boosting notebook and the segmentation/forecasting notebooks are optional practice: their skills are graded inside the final-project milestones instead of as separate homework.
- Framing & baseline (20%) — unit, target, horizon, decision stated; the naive baseline computed and beaten
- Pipeline & leakage (30%) — split before any fitted transformer; leaks identified and removed; code runs top to bottom
- Evaluation (30%) — the right metric for the decision on a held-out set that matches production; results interpreted in plain words
- Communication (20%) — one clear paragraph per question answering what the numbers mean for the business
Midterm Model Competition (teams of 2–3)
20%A Kaggle community competition on a hidden test split of hotel bookings — predict cancellations. Launches Oct 19; leaderboard closes Mon Nov 9, 4:30 PM (upload the same predictions file on Canvas by then); reveal and 3-minute "what worked" talks that evening; one-page memo on Canvas by 11:59 PM.
- Private-leaderboard band (50%) — top third / middle / bottom third on ROC-AUC, with a floor for any leak-free submission that beats the logistic baseline
- One-page modeling memo (50%) — what you tried, what worked, what leaked, and one thing you would do with two more weeks
Final Project — Your Prediction Story (teams of 2–3, solo allowed)
40%A real prediction question, a leak-proof model, an honest evaluation, a decision with a number, and a trust story. Four small pass/fail milestones keep every team moving: problem + dataset (Mon Nov 2), baseline + metric (Mon Nov 23), tuned model + error analysis (Mon Nov 30), dry-run slides (Mon Dec 7, 5:00 PM). Notebook + 2-page memo + repo link Mon Dec 14, 5:00 PM; 8-minute presentation that evening. No written final exam.
- Framing & data (15%)
- Modeling & validation (25%)
- Decision & value (25%)
- Trust story (15%)
- Memo & presentation (20%)
Everything is free
- Google Colab (free) — every notebook runs in the browser; nothing to install
- A Kaggle account with your suffolk.edu email — for the midterm competition and project datasets
- Python 3.12 · pandas · scikit-learn ≥ 1.6 · LightGBM · XGBoost · SHAP · Optuna · MAPIE · statsforecast (all preinstalled or one pip line in Colab; pinned in the repo's requirements.txt for local work)
- AI coding assistants (Colab's built-in assistant, Claude, ChatGPT) — allowed and taught; see the AI policy
Setup walkthrough on the Start Here page. Dataset loaders on the Datasets page.
Course Policies
Attendance
This is an evening seminar that meets once a week — every session matters, and every session has a live lab. More than two unexcused absences will result in a grade reduction. Please notify the instructor in advance if you need to miss a class; the notebook, slides, and explorer for every session are on this site so you can catch up.
AI Tool Usage
AI coding assistants (Claude, ChatGPT, Colab AI, Cursor, Copilot) are encouraged — working data scientists use them every day. Two rules: disclose what you used in a short note at the top of the notebook, and be able to explain and defend every line of a model you submit. A model you cannot explain is a model you cannot trust, and in the presentation the AI will not be standing next to you. Session 12 teaches the verification habit we expect.
Academic Integrity
All submitted work must be your own or properly attributed. Collaboration is encouraged on labs and the team competition; individual homework must reflect individual effort. Copying a Kaggle kernel and changing variable names is not modeling. Suffolk University's academic integrity policy applies in full.
Late Submissions
Assignments are due at 11:59 PM on the due date, on Canvas. Late submissions receive a 10% penalty per day, up to 3 days. After 3 days, assignments receive zero credit. The competition leaderboard closes when it closes. One free 48-hour extension per student on any single homework — just email before the deadline.
Accessibility & Support
If you have a documented disability or need accommodations, please contact Suffolk's Office of Disability Services and let the instructor know early in the semester so arrangements can be made. Students facing personal or financial difficulty are encouraged to reach out — the earlier, the more options.
Communication & Office Hours
Email harslan@suffolk.edu with "ISOM 835" in the subject; expect a reply within one business day. Office hours by appointment before class on Mondays and on Zoom; project weeks get extra slots posted on Canvas. Course announcements go out on Canvas.
This syllabus is a living document; changes are announced on Canvas and reflected here. Last updated Sep 19, 2026.