ML
The ML tab is where Binloops goes beyond rules based trading. Train a personal machine learning model on your own indicator history, then let it make directional predictions alongside your strategies. After logging in head to the ML tab and enable your models to start collecting data.
Collect snapshots
Every time one of your active strategies evaluates a window, Binloops automatically captures a snapshot of every indicator value at that moment — whether a trade fires or not. Once the window resolves, the snapshot is labeled with the actual outcome (UP or DOWN). These become your training data. The more diverse your active strategies, the richer the feature set.
Train your model
Hit Train and the system looks through your snapshot history for patterns like "when RSI was below 35 AND MACD was negative, the market went UP 62% of the time," builds hundreds of these rules, and combines them into a single prediction. Training uses 5-fold time series cross-validation to avoid memorizing patterns that won't repeat.
After training you'll see: accuracy, top contributing features, and a confidence breakdown (e.g. "at 55%+ confidence, 63% accurate on 80 trades"). Select the threshold (confidence) you want your model to trade at and select which assets you want it trading.
Create an ML strategy
Set a confidence threshold (50%–70%). The model runs on every window and generates a signal only when its confidence meets your threshold. Each ML strategy trades BTC, ETH, and SOL independently, the model trains separately per asset. You're limited to 2 ML strategies per account and 3 training runs per hour.
Reading your results:
Accuracy above 55% is meaningful — these are short-term binary markets and anything consistently above coin-flip is an edge
Higher confidence threshold = fewer trades but (usually) better accuracy
Top features show what's driving predictions — if RSI is #1, the model is leaning on momentum signals
Last updated