ML-Ready Forex Datasets — Triple-Barrier Labels, Next-Bar Direction & Time Splits

Available ML-Ready Datasets

Each file includes features, labels and a split column. Downloads are email-gated. Parquet is recommended for Python training.

Label Definitions

Feature Dictionary & Leakage Rules

The feature dictionary lists every column, its type, a description and whether it is a feature, label or metadata. All feature columns are causal. Labels use future bars by design. Descriptive pivot columns were removed from the ML files to prevent accidental leakage.

Minimal Python Example

import pandas as pd

df = pd.read_parquet("XAUUSD_M15_ml_ready.parquet")
features = [c for c in df.columns if c not in {
    "Timestamp", "tb_label", "tb_bars", "tb_return",
    "next_bar_direction", "next_bar_return", "split"
}]
train = df[df.split == "train"]
valid = df[df.split == "validation"]
test  = df[df.split == "test"]

X_train, y_train = train[features], train["tb_label"]
X_valid, y_valid = valid[features], valid["tb_label"]
X_test,  y_test  = test[features],  test["tb_label"]

Encode session and volatility_regime before fitting. Drop warm-up rows with empty indicators. Do not shuffle time-series rows before splitting.

Related Data & Tools