LightGBM
The same idea as Gradient Boosting — trees correcting each other in a chain — but built to train much faster with many data columns at once.
Why it matters
Here, "columns" means every indicator on this site evaluated together. Recent literature (arXiv, not just vendor sources) finds this kind of model competitive with — sometimes better than — LSTM-style neural networks for daily series with a few thousand rows, like this one, without needing the data volume or training cost a neural network does.
Build it yourself
How it works, step by step
- It uses the same sequential-correction mechanism as Gradient Boosting (see the previous model): each new tree learns to correct the error of the ones before it.
- Instead of testing every possible split point of every indicator one by one (the classic method), it buckets values into a small number of "bins" before searching for the best split — much faster, with little loss of accuracy.
- It grows each tree by adding leaves wherever they reduce error the most, rather than level by level — it tends to reach good results with fewer trees overall.
Predictions
These are the real predictions stored day by day, each one produced by walk-forward validation: every day's was produced by a model trained only on prior cycles. They're published for transparency, not because there's an edge behind them — for price direction, the project's research found none that is defensible.
Probability of a rise within 7 days
Probability of a rise within 30 days
Probability of a rise within 90 days
Data last updated on 2026-09-09.