Back to projects

06 / Deep Learning

Multi-output Neural Network

A TensorFlow / Keras network with three outputs: mutation flag, class label and disease risk. Experiments showed strong performance alongside signs of overfitting.

Overview

An experimental neural-network project in Python, TensorFlow and Keras that produces a mutation flag, a class label and a disease-risk output.

Problem

The modelling task has three distinct outputs to learn within one multi-output network.

Architecture

A multi-output model produces three prediction outputs. The diagram describes the output structure, not a verified layer-by-layer architecture.

  1. InputsModel features
  2. NetworkTensorFlow / Keras
  3. 3 outputsMutation / class / risk

Implementation

  • Build a multi-output neural-network experiment with TensorFlow and Keras.
  • Produce mutation-flag, class-label and disease-risk outputs.
  • Examine experimental performance alongside observed signs of overfitting.

Challenges

Strong experimental scores can coexist with overfitting. Each output needs its own evaluation, and generalisation must be established separately.

Technology

  • TensorFlow
  • Keras
  • Python

Outcome

Experiments showed strong performance but also signs of overfitting. These results need further evaluation before drawing conclusions about generalisation.

What I learned

Apparent performance is not the same as generalisation. A multi-output model needs output-specific analysis and careful interpretation of training and validation behavior.

Explore the other projects