Thomas BrochardDevelopment · Data · AI
Let’s talk
Back to selected work

Analytics & ML

Case study / 06

Building Energy Prediction

Connect model exploration to a clear input contract.

Structural feature engineering, comparable regression workflows and a typed prediction contract for building-energy estimates.

Technologies & concepts

  • Python
  • pandas
  • NumPy
  • scikit-learn
  • BentoML
  • Pydantic
  • Jupyter

Technical demonstration / selected source

Private repository

01 / The context

The problem behind the design.

Building-energy estimates depend on consistent inputs and comparable evaluation. The source workflow connects structural feature engineering and regression comparison to a typed prediction interface, using the Seattle 2016 benchmarking schema as reference context.

02 / Engineering decisions

Choices made explicit.

Use structural predictors

Validity filters retain multifamily building observations. Age bands, parking ratios and property-use proportions describe buildings while measured energy quantities are excluded from predictors.

Keep comparisons consistent

A mean baseline, linear regression, SVR and random forest share common folds. Preprocessing stays inside the modelling pipeline to fit within each fold.

Validate the prediction boundary

The API contract checks exact feature names, finite numerical values and categorical types before ordered pipeline prediction.

03 / System view

Follow the flow.

Building-energy modelling design: benchmark-schema inputs are filtered, transformed into structural features, compared across models and connected to a typed prediction contract.

Illustrative architecture from the project presentation. It describes the selected design; it does not imply a deployed service or a measured operational result.

View the full-size diagram
  1. Benchmark schema

  2. Validity filters

  3. Structural features

  4. Model comparison

  5. Parameter search

  6. Prediction contract

04 / Selected artifacts

What the repository contains.

  • source/model_workflow.py
  • model_exploration.ipynb
  • service.py
  • features_config.json

05 / Scope & limitations

How to read this project.

Demonstration and source showcase without reference data, trained models, model stores or a hosted service. The target is weather-normalized site energy use in kBtu. Cross-validation and parameter search are source examples without verified scores. Complete-case selection may bias the sample; a separate holdout, uncertainty assessment and drift checks are needed before relying on predictions. The API expects pre-engineered features and a separately registered model.

Explore the source with access.

The GitHub repository is private. Its link works for authorized collaborators; other visitors can contact me to discuss the project.

Next case study / 07

Education Market Analysis

Analytics & ML

Explore the collection

The next conversation

Let’s talk about your project.

↗
A need, an idea, a technical question.
LinkedIn ↗GitHub ↗