Model Comparison - CryptoView Pro

Quick Comparison

XGBoost

Best for: 1-72 hours
Strength: High short-term accuracy
Speed: Fast training, slow prediction

Prophet

Best for: 1 week - 1 month
Strength: Trend & seasonality
Speed: Slow training, fast prediction

Hybrid

Best for: All horizons
Strength: Automatic selection
Speed: Slow training, fast prediction

Detailed Comparison Table

Performance Characteristics

Feature	XGBoost	Prophet	Hybrid
Optimal Horizon	1-72 hours	1 week - 1 month	All (1h - 1 month)
Short-term Accuracy (1-24h)	⭐⭐⭐⭐⭐ (Best)	⭐⭐ (Poor)	⭐⭐⭐⭐⭐ (Best)
Medium-term Accuracy (1-3d)	⭐⭐⭐⭐ (Good)	⭐⭐⭐ (Fair)	⭐⭐⭐⭐⭐ (Best)
Long-term Accuracy (1w+)	⭐⭐ (Poor)	⭐⭐⭐⭐⭐ (Best)	⭐⭐⭐⭐⭐ (Best)
Typical MAPE (24h)	2-4%	6-10%	2-4%
Typical MAPE (7d)	8-15%	8-12%	8-12%
Direction Accuracy	60-70% (short)	50-58% (long)	Varies by horizon

Technical Specifications

Aspect	XGBoost	Prophet	Hybrid
Algorithm	Gradient Boosting	Additive Decomposition	Ensemble
Training Time	5-15 sec	10-30 sec	15-45 sec
Prediction Time (24h)	200-500ms	50-100ms	50-100ms
Prediction Time (168h)	1-3 sec	50-100ms	50-100ms
Memory Usage	Medium	Low	High
Model Size	~5-20 MB	~1-5 MB	~10-25 MB
Min Data Points	100	100	100
Recommended Data	1000+	1000+	1000+

Feature Capabilities

Feature	XGBoost	Prophet	Hybrid
Feature Engineering	60+ auto features	None (just ds/y)	60+ (XGBoost)
Technical Indicators	✅ RSI, MACD, BB	❌ Not supported	✅ Via XGBoost
Trend Detection	❌ Implicit only	✅ Explicit component	✅ Via Prophet
Seasonality	❌ Via lag features	✅ Daily/weekly	✅ Via Prophet
Confidence Intervals	⚠️ Estimated	✅ Native	✅ Both
Missing Data	❌ Must drop	✅ Handles gaps	⚠️ Mixed
Interpretability	⭐⭐ Feature importance	⭐⭐⭐⭐⭐ Clear components	⭐⭐⭐ Medium

Model Parameters

XGBoost Configuration

XGBoostCryptoPredictor(
    n_estimators=200,        # Number of trees
    learning_rate=0.07,      # Gradient descent step size
    max_depth=6,             # Tree depth
    subsample=0.8,           # Sample fraction per tree
    colsample_bytree=0.8     # Feature fraction per tree
)

Key Parameters:

n_estimators: More trees = better fit, slower training (100-500)
learning_rate: Lower = more conservative, needs more trees (0.01-0.3)
max_depth: Deeper = more complex patterns, risk overfitting (3-10)

Prophet Configuration

ProphetCryptoPredictor(
    changepoint_prior_scale=0.5,  # Trend flexibility
    seasonality_prior_scale=10,   # Seasonality strength
    interval_width=0.95           # Confidence interval width
)

Key Parameters:

changepoint_prior_scale: Higher = more flexible trend (0.001-0.5)
seasonality_prior_scale: Higher = stronger seasonality (1-20)
interval_width: Confidence level for bounds (0.8-0.99)

Hybrid Configuration

HybridCryptoPredictor()
# No parameters - uses default XGBoost and Prophet configs
# Automatically handles model selection and weighting

Model Selection:

≤24h: XGBoost only (100% weight)
24-72h: Weighted ensemble (gradient from XGBoost → Prophet)
72h: Prophet dominant (or 100% Prophet at 168h)

Prediction Comparison

Typical Accuracy by Horizon

1-24 Hours
1-3 Days
1 Week
1 Month

Model	MAPE	Direction Acc	Best Use
XGBoost	2-4%	60-70%	✅ Recommended
Prophet	6-10%	52-58%	❌ Too inaccurate
Hybrid	2-4%	60-70%	✅ Recommended

Winner: XGBoost or Hybrid (uses XGBoost)

Model	MAPE	Direction Acc	Best Use
XGBoost	4-8%	55-65%	✅ Good
Prophet	6-10%	50-60%	⚠️ Fair
Hybrid	3-6%	58-67%	✅ Recommended

Model	MAPE	Direction Acc	Best Use
XGBoost	10-18%	48-54%	❌ Poor
Prophet	8-12%	52-58%	✅ Recommended
Hybrid	8-12%	52-58%	✅ Recommended

Model	MAPE	Direction Acc	Best Use
XGBoost	15-30%	45-50%	❌ Unusable
Prophet	12-20%	48-55%	✅ Recommended
Hybrid	12-20%	48-55%	✅ Recommended

Training Comparison

Training Process

predictor = XGBoostCryptoPredictor()

# 1. Creates 60+ features
# 2. Splits train/test (80/20)
# 3. Scales features
# 4. Trains gradient boosting
metrics = predictor.train(df, train_size=0.8)

# Returns: train/test MAE, RMSE, MAPE, direction accuracy

Training Speed Benchmark

Test: 1000 hourly data points, Bitcoin

Model	Training Time	Prediction Time (24h)	Prediction Time (168h)
XGBoost	~8 seconds	~300ms (iterative)	~2 seconds
Prophet	~18 seconds	~60ms (direct)	~80ms
Hybrid	~26 seconds	~60ms	~80ms

Prophet is slower to train but much faster for multi-step predictions because it predicts all periods directly (no iteration).

Prediction Speed Comparison

Why XGBoost is Slower for Multi-step

# XGBoost: ITERATIVE (slow for many periods)
for i in range(24):
    features = create_features(df)  # ~5ms
    prediction = model.predict()     # ~3ms
    df = append(df, prediction)      # Update for next iteration
# Total: 24 * 8ms = ~192ms + overhead

# Prophet: DIRECT (fast for many periods)
future = make_future_dataframe(periods=24)
predictions = model.predict(future)  # ~60ms for all 24
# Total: ~60ms regardless of periods

Prediction Time Chart

Periods	XGBoost	Prophet	Hybrid
1	~10ms	~50ms	~50ms
12	~100ms	~55ms	~55ms
24	~200ms	~60ms	~60ms
72	~600ms	~70ms	~70ms
168	~1.5s	~80ms	~80ms
720	~6s	~120ms	~120ms

For predictions >72 hours, Prophet is 10-50x faster than XGBoost.

Feature Engineering

XGBoost Features (60+)

Returns (4 features)

1-hour return: df['close'].pct_change(1)
4-hour return: df['close'].pct_change(4)
24-hour return: df['close'].pct_change(24)
7-day return: df['close'].pct_change(168)

Moving Averages (12 features)

MA periods: 7, 14, 30, 50
EMA periods: 12, 26, 50
Price-to-MA ratios for each MA

Volatility (3 features)

Rolling std: 7, 14, 30 periods
Based on 1-hour returns

Bollinger Bands (4 features)

Middle band (20-period MA)
Upper band (middle + 2*std)
Lower band (middle - 2*std)
Band position (normalized)

Technical Indicators (6+ features)

RSI normalized, overbought/oversold
MACD difference and signal
High/low ratio
Close/open ratio

Volume (3 features)

Volume MA (7 periods)
Volume ratio
Volume change

Temporal (4 features)

Hour of day (0-23)
Day of week (0-6)
Day of month (1-31)
Month (1-12)

Lag Features (5 features)

Close lag: 1, 2, 3, 7, 14 periods

Prophet Features (2)

ds: Timestamp (automatically extracts seasonality)
y: Close price

Derived internally:

Trend component
Daily seasonality (24-hour patterns)
Weekly seasonality (7-day patterns)

Use Case Recommendations

Choose XGBoost When:

✅ Prediction horizon is 1-72 hours
✅ You have technical indicators (RSI, MACD, etc.)
✅ Need highest short-term accuracy
✅ Intraday trading strategies
✅ Rich OHLCV historical data available
✅ Care about direction accuracy >60%

Choose Prophet When:

✅ Prediction horizon is >1 week
✅ Need trend decomposition
✅ Want native confidence intervals
✅ Weekly/monthly planning
✅ Seasonality analysis is important
✅ Have missing data gaps
✅ Need fast multi-step predictions

Choose Hybrid When:

✅ Prediction horizon varies (1h to 1 month)
✅ Want automatic model selection
✅ Building a general-purpose forecasting system
✅ Need best-of-both-worlds approach
✅ Production deployment
✅ Don’t want to manually choose models
✅ Value smooth transitions between models

Code Examples

Quick Start Comparison

from models.xgboost_model import XGBoostCryptoPredictor

predictor = XGBoostCryptoPredictor()
predictor.train(df, train_size=0.8)
predictions = predictor.predict_future(df, periods=24)

print(predictions['predicted_price'])

Backtesting Comparison

from models.xgboost_model import backtest_model, XGBoostCryptoPredictor
from models.prophet_model import backtest_prophet, ProphetCryptoPredictor

# XGBoost backtest
xgb_predictor = XGBoostCryptoPredictor()
xgb_results = backtest_model(df, xgb_predictor, train_size=0.8)
print(f"XGBoost Test MAPE: {xgb_results['metrics']['test_mape']:.2f}%")

# Prophet backtest
prophet_predictor = ProphetCryptoPredictor()
prophet_results = backtest_prophet(df, prophet_predictor, test_periods=168)
print(f"Prophet Test MAPE: {prophet_results['test_metrics']['mape']:.2f}%")

# Compare direction accuracy
print(f"XGBoost Direction: {xgb_results['metrics']['test_direction_accuracy']:.2f}%")
print(f"Prophet Direction: {prophet_results['test_metrics']['direction_accuracy']:.2f}%")

Decision Flowchart

Summary Table

Criteria	XGBoost	Prophet	Hybrid
Best Horizon	1-72h	1w-1m	All
Accuracy (Short)	⭐⭐⭐⭐⭐	⭐⭐	⭐⭐⭐⭐⭐
Accuracy (Long)	⭐⭐	⭐⭐⭐⭐⭐	⭐⭐⭐⭐⭐
Training Speed	⭐⭐⭐⭐	⭐⭐⭐	⭐⭐
Prediction Speed	⭐⭐	⭐⭐⭐⭐⭐	⭐⭐⭐⭐⭐
Interpretability	⭐⭐	⭐⭐⭐⭐⭐	⭐⭐⭐
Ease of Use	⭐⭐⭐	⭐⭐⭐⭐	⭐⭐⭐⭐⭐
Memory Usage	⭐⭐⭐	⭐⭐⭐⭐⭐	⭐⭐
Confidence Intervals	⭐⭐⭐	⭐⭐⭐⭐⭐	⭐⭐⭐⭐⭐
Feature Engineering	Auto (60+)	None needed	Auto (60+)
Data Requirements	1000+	1000+	1000+
Production Ready	✅	✅	✅

Final Recommendations

For Most Users

Use the Hybrid model - it automatically selects the best approach for your horizon and provides optimal predictions across all timeframes.

For High-Frequency Traders

Use XGBoost - maximum short-term accuracy for intraday strategies.

For Strategic Planning

Use Prophet - clear trend analysis and seasonality for long-term decisions.

Next Steps

XGBoost Details

Deep dive into XGBoost model

Prophet Details

Deep dive into Prophet model

Hybrid Details

Deep dive into Hybrid model

Get Started

Core Concepts

ML Models

Features

Configuration

​Quick Comparison

XGBoost

Prophet

Hybrid

​Detailed Comparison Table

​Performance Characteristics

​Technical Specifications

​Feature Capabilities

​Model Parameters

​XGBoost Configuration

​Prophet Configuration

​Hybrid Configuration

​Prediction Comparison

​Typical Accuracy by Horizon

​Training Comparison

​Training Process

​Training Speed Benchmark

​Prediction Speed Comparison

​Why XGBoost is Slower for Multi-step

​Prediction Time Chart

​Feature Engineering

​XGBoost Features (60+)

​Prophet Features (2)

​Use Case Recommendations

​Choose XGBoost When:

​Choose Prophet When:

​Choose Hybrid When:

​Code Examples

​Quick Start Comparison

​Backtesting Comparison

​Decision Flowchart

​Summary Table

​Final Recommendations

For Most Users

For High-Frequency Traders

For Strategic Planning

​Next Steps

XGBoost Details

Prophet Details

Hybrid Details

Build docs developers (and LLMs) love

Quick Comparison

Detailed Comparison Table

Performance Characteristics

Technical Specifications

Feature Capabilities

Model Parameters

XGBoost Configuration

Prophet Configuration

Hybrid Configuration

Prediction Comparison

Typical Accuracy by Horizon

Training Comparison

Training Process

Training Speed Benchmark

Prediction Speed Comparison

Why XGBoost is Slower for Multi-step

Prediction Time Chart

Feature Engineering

XGBoost Features (60+)

Prophet Features (2)

Use Case Recommendations

Choose XGBoost When:

Choose Prophet When:

Choose Hybrid When:

Code Examples

Quick Start Comparison

Backtesting Comparison

Decision Flowchart

Summary Table

Final Recommendations

Next Steps