Case study · AI & Data Science Program · MIT IDSS · 2026

Applied Machine Learning: MIT IDSS Program Projects

Three business problems, three kinds of machine learning, each ending in a decision a business could act on.

  • 922 / 1,318customers in two segments
  • 85%of converting leads found
  • 0.8808test RMSE, tuned SVD
  • Mar–Jul 2026MIT IDSS program
On this page

The program

AI & Data Science Program: Leveraging Responsible AI, Data and Statistics for Practical Impact, MIT IDSS (Institute for Data, Systems, and Society), delivered through Great Learning, March to July 2026; certificate issued 15 July 2026. Verify the certificate (external site).

It added the statistics and model evaluation behind supervised, unsupervised and recommender models to my earlier applied AI work (image similarity search with vector embeddings).

The projects

Project Decision it supports Headline result
Customer segmentation (K-means) Who gets premium offers, who gets discounts? Two segments of 922 and 1,318 customers with opposite spending and channel habits
Lead-conversion prediction (decision tree, random forest) Which leads should sales call first? On held-out test leads, the tuned forest finds 85% of converters, at 83% accuracy
Product recommender (KNN, SVD) What should each shopper see next? Tuned SVD: test RMSE 0.8808 vs. 0.9526 for the best KNN; its F1@10 (0.866) is flattered by high average ratings

Every project compares at least two approaches, ends with an action a team could take on Monday, and has a short, labelled design for running it on Databricks.

Decisions

Publish my own write-ups, not the course notebooks. Context: the notebooks contain course-provided problem statements and hints; the work and results are mine. Options: link the notebooks; publish cleaned repos with my own write-up; case studies only. Decision: cleaned repos plus case studies, once reviewed. Consequences: more rewriting; respects the course’s terms. No grades are published.

Projects

  • Unsupervised learning (K-means clustering)

    922 / 1,318customers in two segments

    Customer Segmentation with K-means

    K-means on 2,240 retail customers: two segments a marketer can act on.

  • Classification (decision tree, random forest)

    85%of converting leads found

    Lead-Conversion Prediction

    Decision tree and random forest on 4,612 leads, tuned for recall.

  • Recommender systems (popularity, KNN collaborative filtering, SVD matrix factorization)

    0.8808test RMSE, tuned SVD

    Product Recommendation System

    Four recommenders compared on 65,290 ratings, and a metric blind spot that changed the design.

Next case study

  • AI & Data Science Program · Project 1

    922 / 1,318customers in two segments

    Customer segmentation

    K-means on 2,240 retail customers: two segments a marketer can act on.

Need a Claude agent or data workflow built? Work with me