Crypto Historical Data Retrieval

A Python project that retrieves historical OHLC trading data for top cryptocurrency pairs, computes analytical trading metrics, exports results to Excel, and trains a machine learning model to predict future price movements.

Features

• Fetches daily OHLC data from the Binance public API (no API key required)
• Computes configurable look-back and look-forward price metrics
• Exports data to a multi-sheet Excel workbook
• Trains a multi-output regression model to predict future price differences

Project Structure

.
├── crypto_data.py              # Core: fetch_crypto_data(), calculate_metrics()
├── ml_model.py                 # ML: train_model(), predict_outcomes()
├── main.py                     # Orchestration script
├── crypto_historical_data.xlsx # Generated Excel output
└── Project_Explanation.txt     # Approach, decisions, challenges

Setup

pip install pandas requests openpyxl scikit-learn numpy

Usage

Run the full pipeline

python main.py

This fetches data for 5 pairs, computes metrics, saves the Excel workbook, trains the ML model, and runs a sample prediction.

Use modules individually

from crypto_data import fetch_crypto_data, calculate_metrics

# Fetch daily OHLC data
df = fetch_crypto_data("BTC/USDT", "2023-01-01")

# Compute metrics (7-day lookback, 5-day lookahead)
df = calculate_metrics(df, variable1=7, variable2=5)

from ml_model import train_model, predict_outcomes

bundle = train_model(df, variable1=7, variable2=5)

prediction = predict_outcomes(
    bundle,
    days_since_high=1,
    pct_diff_from_high=-0.95,
    days_since_low=3,
    pct_diff_from_low=2.50,
)
# {'%_Diff_From_High_Next_5_Days': ..., '%_Diff_From_Low_Next_5_Days': ...}

API Details

Property	Value
Provider	Binance Public REST API
Endpoint	https://api.binance.com/api/v3/klines
Auth	None (public endpoint)
Pairs supported	1000+ (USDT, BTC, ETH quoted)
Timeframes	1m, 5m, 15m, 1h, 4h, 1d, 1w, 1M
Data from	Each asset's listing date (BTC: 2017)
Rate limit	1 200 requests/min

Calculated Metrics

Given variable1 (look-back days) and variable2 (look-forward days):

Column	Description
High_Last_{v1}_Days	Rolling max of High over past v1 days
Days_Since_High_Last_{v1}_Days	Days since that high occurred
%_Diff_From_High_Last_{v1}_Days	(Close − High_Last) / High_Last × 100
Low_Last_{v1}_Days	Rolling min of Low over past v1 days
Days_Since_Low_Last_{v1}_Days	Days since that low occurred
%_Diff_From_Low_Last_{v1}_Days	(Close − Low_Last) / Low_Last × 100
High_Next_{v2}_Days	Max of High over next v2 days
%_Diff_From_High_Next_{v2}_Days	(High_Next − Close) / Close × 100
Low_Next_{v2}_Days	Min of Low over next v2 days
%_Diff_From_Low_Next_{v2}_Days	(Low_Next − Close) / Close × 100

ML Model

Type: Multi-output regression (predicts two targets simultaneously)

Features:

• Days_Since_High_Last_{v1}_Days
• %_Diff_From_High_Last_{v1}_Days
• Days_Since_Low_Last_{v1}_Days
• %_Diff_From_Low_Last_{v1}_Days

Targets:

• %_Diff_From_High_Next_{v2}_Days
• %_Diff_From_Low_Next_{v2}_Days

Two models are trained (RandomForest and GradientBoosting); the one with lower test RMSE is selected. Features are scaled with StandardScaler. Data is split chronologically (80/20) to respect time-series ordering.

Excel Output

The workbook crypto_historical_data.xlsx contains:

• One sheet per crypto pair (BTC_USDT, ETH_USDT, etc.)
• A Combined sheet with all pairs merged (used for ML training)