Topfolio

Navigation

Projects/Demand Forecasting with Machine Learning
advanced5 milestones · ~15 hours

Demand Forecasting with
Machine Learning

An operations director needs weekly demand forecasts to plan inventory. You'll explore historical order patterns, engineer time-series features (lags, rolling averages, seasonality), train and evaluate models against honest holdouts, and ship a portfolio piece that balances analytical rigor with production-thinking. Advanced — assumes pandas + basic ML comfort.

What you'll learn

  • —Frame a business question (weekly inventory planning) as an ML forecasting problem
  • —Engineer time-series features (lags, rolling stats, seasonality flags) without leaking the future into the past
  • —Compare baseline and ML models honestly using time-series cross-validation
  • —Reason about model deployment: cadence, monitoring, when to retrain
  • —Ship a portfolio repo that demonstrates both analytical rigor and production-thinking

After this project

What you'll be able to claim — credibly — once your repo is shipped.

Roles you can credibly apply to

Junior Data Scientist (time-series)
End-to-end forecasting — from problem framing to honest evaluation to deployment thinking — is the exact loop hiring managers test for in junior DS roles.
— Pair with at least one classification or causal-inference project to show you're not a 'forecasting only' candidate.
ML Engineer (entry-level)
The production-thinking section (cadence, monitoring, retraining) is what separates 'student model' from 'production-ready' work — directly relevant to MLE roles.
— MLE roles often expect engineering rigor (Docker, CI, model registries); this project shows the conceptual layer, not the infra layer.
Forecasting / Demand Planning Analyst
Operations + supply-chain teams hire specifically for demand forecasting — domain experience with weekly horizons and seasonal patterns transfers directly.
— Industry-specific (retail, CPG, logistics) signal helps; this is an ecommerce framing.

Keywords on your resume after this

Tools
Pythonscikit-learnpandasNumPyJupytermatplotlibPostgreSQLGit
Methods
Time-series forecastingFeature engineering (lags, rolling stats)Time-series cross-validationSeasonal decompositionBaseline modelingModel evaluation (RMSE, MAPE)
Concepts
Demand forecastingTemporal hygiene (no leakage)Production deploymentModel monitoringRetraining cadenceFeature importance interpretation

Interview questions you'll be ready for

  1. 01Walk me through how you'd frame 'I want better demand forecasts' as a concrete ML problem — including what metric you'd optimize.
  2. 02What's the single most common mistake in time-series feature engineering, and how do you prevent it?
  3. 03Your seasonal-naive baseline beats your fancy XGBoost by 5%. What does that tell you?
  4. 04How would you set up time-series cross-validation differently from random k-fold, and why does it matter?
  5. 05If your model's deployed and starts drifting, what would your monitoring actually catch — and what would it miss?
Resume-ready bullets — unlocked when you reach Milestone 5
After completing the project, you'll get a copy-ready resume bullet template phrased the way hiring managers want to hear it, with the numbers your own analysis produced.

The 5 milestones

Each milestone is reviewed before you advance.

  1. 01

    Set up your project repo

    Before any analytical work: create a public GitHub repository, push the standard skeleton, and paste the URL into your workspace.

  2. 02

    Frame the problem and explore the demand data

    Translate the operations ask into a forecasting problem and form a first-pass mental model of the demand patterns.

  3. 03

    Engineer time-series features

    Build features from order history without leaking the future into the past.

  4. 04

    Train, evaluate, compare

    Build a baseline, build an ML model, evaluate both honestly with time-series cross-validation.

  5. 05

    Production thinking and portfolio

    Wire the analysis into a deployment-ready story and polish the repo for a recruiter.

Reading & references

Pre-flight setup: Python ML environment
Setup Guide · Premium
Time series 101 — what to look for first
Article
Feature engineering for time series — without leakage
Article · Premium
Time-series cross-validation, properly
Article
Why you need a baseline (and what makes a good one)
Article
ML in production: monitoring + retraining patterns
Article · Premium
Anatomy of a great portfolio README
Article

Build your portfolio & get certified

Want more projects like this? Join our Data Analyst Work Experience Program to complete 5+ guided industry projects, gain real experience, and earn an internship certificate.

Join the Data Analyst Work Experience Program