AI dan Perdagangan: Cara Menggunakan Pembelajaran Mesin untuk Perdagangan Otomatis Cryptocurrency di 2026

Berbasis Aturan (Rule-Based):
IF RSI < 30 → Beli
IF RSI > 70 → Jual

Kelebihan: Sederhana, dapat diinterpretasikan, dapat diprediksi
Kekurangan: Tidak dapat beradaptasi dengan perubahan pasar, hanya menangkap pola sederhana

Berbasis Data (Data-Driven):
INPUT: Harga, volume, indikator teknis, sentimen berita, data on-chain
MODEL: Belajar pola tersembunyi dari puluhan hingga ribuan variabel
OUTPUT: Probabilitas beli/jual/tahan dan perkiraan pengembalian

Kelebihan: Menangkap pola kompleks, dapat beradaptasi dengan beberapa perubahan pasar
Kekurangan: Black box, risiko overfitting, memerlukan banyak data

pip install pandas numpy scikit-learn xgboost ta ccxt matplotlib

import pandas as pd
import numpy as np
from ta import add_all_ta_features

def create_features(df: pd.DataFrame) -> pd.DataFrame:
    """Membuat fitur ML dari data OHLCV"""

    # Indikator teknis dasar (menggunakan pustaka ta)
    df = add_all_ta_features(
        df, open="open", high="high", low="low",
        close="close", volume="volume", fillna=True
    )

    # Fitur perubahan harga
    for period in [1, 3, 7, 14, 30]:
        df[f'return_{period}d'] = df['close'].pct_change(period)

    # Fitur volatilitas
    df['volatility_7d'] = df['close'].pct_change().rolling(7).std()
    df['volatility_30d'] = df['close'].pct_change().rolling(30).std()

    # Deteksi anomali volume
    df['volume_ratio'] = df['volume'] / df['volume'].rolling(20).mean()

    # Variabel target: Jika harga 4 jam ke depan naik lebih dari 1%, maka 1, jika tidak 0
    df['target'] = (df['close'].shift(-4) > df['close'] * 1.01).astype(int)

    return df.dropna()

from xgboost import XGBClassifier
from sklearn.model_selection import TimeSeriesSplit
from sklearn.metrics import classification_report

def train_model(df: pd.DataFrame):
    # Memisahkan fitur dan target
    exclude_cols = ['open', 'high', 'low', 'close', 'volume', 'target']
    feature_cols = [c for c in df.columns if c not in exclude_cols]

    X = df[feature_cols]
    y = df['target']

    # Validasi silang deret waktu (untuk mencegah kebocoran data masa depan)
    tscv = TimeSeriesSplit(n_splits=5)

    model = XGBClassifier(
        n_estimators=200,
        max_depth=5,
        learning_rate=0.05,
        subsample=0.8,
        colsample_bytree=0.8,
        random_state=42
    )

    # Menggunakan 20% terakhir sebagai set pengujian
    split = int(len(X) * 0.8)
    X_train, X_test = X[:split], X[split:]
    y_train, y_test = y[:split], y[split:]

    model.fit(X_train, y_train)

    # Evaluasi kinerja
    y_pred = model.predict(X_test)
    print(classification_report(y_test, y_pred))

    return model

# Contoh penggunaan
# model = train_model(df_with_features)

def ml_trading_signal(model, current_features: pd.DataFrame) -> str:
    """Menghasilkan sinyal perdagangan menggunakan model ML"""

    proba = model.predict_proba(current_features)[0]
    buy_prob = proba[1]   # Probabilitas kenaikan

    # Sinyal berbasis probabilitas (berhati-hati saat tidak pasti)
    if buy_prob > 0.65:
        return 'BUY'
    elif buy_prob < 0.35:
        return 'SELL'
    else:
        return 'HOLD'

import ccxt
import pandas as pd
from datetime import datetime, timedelta

def fetch_historical_data(symbol: str, timeframe: str, days: int) -> pd.DataFrame:
    """Mengumpulkan data historis dari Bithumb"""

    exchange = ccxt.bithumb()
    since = exchange.parse8601(
        (datetime.now() - timedelta(days=days)).strftime('%Y-%m-%dT%H:%M:%S')
    )

    all_candles = []
    while since < exchange.milliseconds():
        candles = exchange.fetch_ohlcv(symbol, timeframe, since=since, limit=200)
        if not candles:
            break
        all_candles.extend(candles)
        since = candles[-1][0] + 1

    df = pd.DataFrame(all_candles, columns=['timestamp', 'open', 'high', 'low', 'close', 'volume'])
    df['timestamp'] = pd.to_datetime(df['timestamp'], unit='ms')
    return df