Multiple Horizons Ensemble Method (MHEMe)

This repository contains the material for the final project of the Statistical Methods course (2025/2026).

Team

• Bortolussi Lorenzo
• Bredariol Francesco
• Savorgnan Enrico

Project Overview

The goal of this project is to develop MHEMe (Multiple Horizons Ensemble Method), a novel ensemble approach for time series forecasting aimed at reducing prediction variance.

The method combines forecasts generated at multiple prediction horizons into a single ensemble prediction, leveraging horizon diversity to improve stability and robustness.

Repository Structure

├─ data             # data used for evaluation of the method
├─ experiments      # folder containing different tests
├─ report           # some reports
├─ results          # folder containing results over benchmarks
├─ literature       # folder containing literature research about the project
├─ src              # method's Python implementation
└─ models           # folder containing models' saving files

Results

The proposed method is evaluated on standard time series datasets and compared against baseline forecasting models in terms of variance and predictive accuracy.

Quick setup

To run the code contained in this repository, please follow the instructions below.

    python -m venv .venv
    .\.venv\Scripts\activate
    pip install -r requirements.txt
    pip install -e .
    nbdime config-git --enable
    git lfs install
    git lfs pull

Note If you are having trouble with some files showing up as git lfs pointer, just copy and paste the following inastructions into the terminal:

    git lfs install
    git lfs pull

License

MIT License