Best approach for football match prediction using TensorFlow?

TensorFlow
1

Hi everyone,

I’m working on a machine learning project focused on predicting football (soccer) match outcomes, and I’d like to get input from the TensorFlow community on best practices and model choice.

The goal is to predict match results (win/draw/loss) or expected goals using historical match data such as:

  • Team performance over time
  • Home vs away
  • Goals scored/conceded
  • Recent form and basic statistics

I’m considering different approaches, including:

  • Classical statistical models (Poisson, Elo-based features)
  • Traditional ML models (logistic regression, Random Forest, XGBoost)
  • Deep learning approaches using TensorFlow (e.g. feedforward networks, LSTM for time series)

From your experience:

  • Does deep learning (TensorFlow/Keras) actually outperform simpler statistical or ML models for this type of problem?
  • Are LSTMs or temporal models worth it given the relatively limited and noisy nature of football data?
  • Any recommended architectures, loss functions, or evaluation metrics for this task?

I’m not looking for betting advice, but rather for a technically sound ML discussion about what works best in practice.

Thanks in advance for any insights or references.

1 Like