Predicting Cryptocurrency Prices with Machine Learning and the CoinGecko API

predictor.py

from flask import Flask, jsonify, render_template, request
import requests
import json
import numpy as np
from sklearn.linear_model import LinearRegression
import time
from sklearn.preprocessing import MinMaxScaler
from datetime import datetime, timedelta

app = Flask(__name__)

API_KEY = "YOUR_API_KEY"

@app.route('/', methods=['GET', 'POST'])
def index():
    # Get the list of available coins
    response = requests.get(f'https://api.coingecko.com/api/v3/coins/markets?vs_currency=usd&api_key={API_KEY}')
    coins_list = response.json()
    coins = [coin['id'] for coin in coins_list]

    if request.method == 'POST':
        coin_id = request.form.get('coin_id')
        days = request.form.get('days')
        if coin_id in coins:
            predictions = get_market_chart(coin_id, days)
            return render_template('index.html', predictions=predictions, coins=coins, coin_id=coin_id, days=days)
        else:
            return render_template('index.html', error="Invalid coin ID", coins=coins)

    return render_template('index.html', coins=coins)

def get_market_chart(coin_id, days):
    # Get the current timestamp in seconds
    end_timestamp = int(time.time())
    # Calculate the start timestamp based on the number of days
    start_timestamp = end_timestamp - int(days) * 24 * 60 * 60

    # Get OHLC data
    response = requests.get(f'https://api.coingecko.com/api/v3/coins/{coin_id}/ohlc?vs_currency=usd&days={days}&api_key={API_KEY}')
    ohlc_data = response.json()

    # Get historical data for a specific date
    date = time.strftime('%d-%m-%Y', time.gmtime(start_timestamp))
    response = requests.get(f'https://api.coingecko.com/api/v3/coins/{coin_id}/history?date={date}&api_key={API_KEY}')
    history_data = response.json()

    # Get current market data
    response = requests.get(f'https://api.coingecko.com/api/v3/coins/markets?vs_currency=usd&ids={coin_id}&api_key={API_KEY}')
    market_data = response.json()[0]

    # Assuming 'ohlc_data' is a list of [time, open, high, low, close] lists
    ohlc = np.array(ohlc_data)

    # Use only the 'time' and 'close' columns
    X = ohlc[:, [0, 4]] # Time and Close price
    y = ohlc[:, 4] # Close price

    # Add historical price and current market data as features
    X = np.concatenate((X, np.full((len(X), 1), history_data['market_data']['current_price']['usd']), np.full((len(X), 1), market_data['current_price'])), axis=1)

    # Scale the data
    scaler = MinMaxScaler()
    X = scaler.fit_transform(X)

    # Train the model
    model = LinearRegression()  
    model.fit(X, y)

    # Make a prediction
    y_pred = model.predict(X)

    # Create a list of dates for the predictions
    start_date = datetime.now()
    dates = [(start_date + timedelta(days=i)).strftime('%Y-%m-%d') for i in range(len(y_pred))]

    # Combine the dates and predictions into a dictionary
    predictions = {date: pred for date, pred in zip(dates, y_pred)}

    return predictions

if __name__ == '__main__':
    app.run(debug=True)