Machine Learning / Predictive application

Broiler FCR Prediction

A hybrid TCN and XGBoost workflow for predicting broiler feed conversion, with explainable results and a Streamlit interface.

View repository Read the documentation

Actual versus predicted FCR scatter plot exported from the project — Training-set prediction plot exported from the repository; not a held-out performance claim.

The context

What this project explores.

Feed Conversion Ratio measures how efficiently broilers turn feed into body weight. This project combines historical farm data, a temporal encoder, and a prediction model to support performance analysis.

The approach

From problem to output.

Prepare farm data

Normalize historical records and engineer features across production periods.

Train a hybrid model

Use a Temporal Convolutional Network as a feature encoder and XGBoost for prediction, with Optuna for parameter search.

Make results usable

Explore predictions through Streamlit and interpret model behavior with SHAP and LIME.

Inside the repository

The deliverables.

Data preparation and feature engineering pipeline
Hybrid model training and evaluation outputs
Streamlit application and integration API code

More to explore.

Back to the collection