Overview

Project Focus:

This project focuses on analyzing the price movements of MKR, the governance token of MakerDAO, using machine learning models.
By identifying patterns and predicting the 1h close, I aim to provide valuable insights for investors and participants in the decentralized finance (DeFi) ecosystem.

MKR Overview:

MakerDAO is a decentralized platform that enables users to generate DAI, a stablecoin pegged to the US dollar, by locking cryptocurrency as collateral.
Unlike traditional financial systems, MakerDAO operates without a central authority, relying on smart contracts on the blockchain.
MKR is the governance token of MakerDAO, meaning that its holders can vote on key decisions, such as risk management policies and system upgrades. MKR’s price fluctuates based on market demand, governance changes, and overall crypto market trends.

DAI Importance:

• Stability is crucial for decentralized finance (DeFi) applications
• DAI is created by locking collateral in MakerDAO smart contracts

Value of Predicting MKR Price:

• MKR impacts the health of DAI and MakerDAO's protocol
• Accurate predictions provide insights into market sentiment, governance decisions, and DAI stability
• Beneficial for DeFi participants and investors

Prediction Target (y):

• y represents the closing price of MKR over consecutive time periods
• It is calculated as:

y = 'close'

• Continuous variable representing the close in MKR price

Importance of Analyzing Close:

• Reveals patterns in MKR's volatility and behavior
• Highlights MKR's impact on DAI stability and the MakerDAO ecosystem
• Provides actionable insights for DeFi participants and investors

Features

Crypto Data Fetcher:

• Retrieves OHLC data for selected cryptocurrencies and stablecoins using the Binance and Kraken API
• Includes additional derived metrics and timezone conversion

Feature Engineering:

• Creates various technical features and custom calculations
• Utilizes the TA-Lib library for advanced technical analysis

MKRUSDT Analysis:

• Focuses on the governance token MKR
• Examines factors influencing its price growth
• Uses machine learning models to predict the target variable (y)

Machine Learning Models:

• Implements models such as Linear Regression (LR), Decision Trees (DT), Random Forest (RF), and XGBoost
• Designed to predict MKR price trends

Flask:

• Included for programmatic interaction with the data
• Optional, suitable for deployment

Docker Support:

• A Dockerfile is provided for easy deployment in containerized environments

Datasets

Crypto Data:
Link to Cryptocurrencies Binance Dataset

Link to Cryptocurrencies Kraken Dataset

Merged Data with Features:
Link to merged Dataset

Structure

stable_coin/  
├── images/                                     # Contains the images that are generated     
│   ├── boxplot_mkr.png      
│   ├── correlation_matrix_mkr.png   
│   ├── cross_validation_mse_vs_max_depth_dt.png
│   ├── distribution_close.png
│   ├── feature_importance_on_accuracy_drop.png 
│   ├── feature_importance_xgboost.png 
│   ├── heatmap_dt.png 
│   ├── mse_heatmap_rf.png 
│   ├── mse_vs_max_depth_dt.png 
│   ├── mse_vs_num_trees_diff_max_depth_rf.png 
│   ├── mse_vs_num_trees_diff_min_samples_rf.png  
│   ├── mse_vs_num_trees_rf.png  
│   ├── predicted_values_distribution_lr.png  
│   ├── price_change_correlation_with_volume.png   
│   ├── residuals_distribution_lr.png  
│   ├── residuals_rf.png   
│   ├── residuals_vs_predicted.png 
│   ├── rmse_vs_max_depth_for_diff_eta.png   
│   ├── scatter_actual_vs_predicted_rf.png 
│   ├── shap_bar_plot_xgboost.png 
│   ├── shap_beeswarm_plot_xgboost.png     
│   ├── timeseries_mkrusdt.png     
│   ├── timeseries_daiusdt.png  
│   ├── timeseries_eur.png    
│   ├── y_boxplot.png 
│   ├── y_histogram.png 
├── README.md                      
├── notebook.ipynb/       
│   ├── get_coins                               # Fetches and processes cryptocurrency data       
│   ├── EDA                                     # EDA on MKRUSDT
│   ├── feature engineering                     # Adds derived metrics for ML models      
│   ├── model evaluation and tuning             # Compares models and saves the best as a pickle file      
├── train.py                                    # Trains the best model            
├── predict.py                                  # Flask application for making predictions       
├── requirements.txt                            # List of required Python packages       
├── environment.yml                             # Conda environment file     
├── LICENSE      
├── Dockerfile                                  # For containerized deployment    

Statistical Summary of MKR Price Data:

Price Trends:

• The average closing price is $1511.79, with a high of $2411.00 and a low of $880.00
• The 7-day and 30-day moving averages are nearly identical, suggesting a stable trend over the period

Volatility & Price Changes:

• 7-day volatility is relatively high (Mean: 12.72, Max: 88.32), indicating price fluctuations
• Intraday price movements are small on average (1.29%), but can spike up to 11.25%
• Price change distribution shows a slight negative bias (Mean: -0.0139), with extreme moves ranging from -4.35% to +6.61%

Market Momentum & Strength:

• ADX (Average Directional Index) is 27.86, indicating moderate trend strength, but can reach as high as 82.55
• RSI (Relative Strength Index) is 48.70 on average, suggesting neutral momentum, but it ranges from 15.72 (oversold) to 97.19 (overbought)

MACD Analysis:

• The MACD mean is slightly positive (1.06), but its large standard deviation suggests high variability
• The MACD histogram shows an overall negative bias (-0.55), indicating weak momentum

Short- & Long-Term Growth:

• 1h and 4h growth rates hover around neutral (≈1.00), with minor fluctuations
• 72h growth shows greater variation (0.7896 to 1.4076), indicating potential medium-term trends

Correlation for MKRUSDT

Key observations:

• Price Change Independence: price_change shows minimal correlation with all other features (values between -0.01 and 0.04), indicating it's relatively independent of other market indicators
• High Multicollinearity Group: There's extremely high correlation (≈1.0) between close, open, and 7d_ma, suggesting redundant information that could lead to model instability
• Medium Correlation Group: atr shows moderate to strong correlation (0.59-0.62) with price levels (close, open, 7d_ma, 30d_ma), indicating its close relationship with recent price action
• Technical Indicator Relationships: rsi has notable correlation with growth metrics (growth_4h: 0.58, growth_72h: 0.54), suggesting these features capture similar market momentum patterns
• Volume Metrics: volume_change shows minimal correlation with other features, making it a potentially valuable independent signal

Boxplot for Closing Prices for MKRUSDT

Key observations:

• Interquartile range (IQR) spans from Q1 = 1288.0 to Q3 = 1668.0, representing the middle 50% of closing prices
• Median close price appears to be around the center of the IQR, indicating a relatively symmetric distribution
• Presence of outliers beyond the upper whisker (above ≈ 2200) suggests occasional price spikes
• Lower whisker extends towards ≈ 900-1000, indicating a historical lower price range
• Distribution suggests that while the majority of prices are within a stable range, significant upward price jumps have occurred, which could be relevant for volatility analysis

Timeseries for MKRUSDT and DAIUSD

Key observations for MKRUSDT:

• Over the last 180 days, MKRUSDT exhibited high volatility
• Price peaked in December 2024, surpassing 2400 USDT, followed by a strong decline
• Continuous downtrend can be observed from January 2025, stabilizing around 1000 USDT in February
• This suggests a period of strong speculative movement, followed by a bearish correction

Key observations for DAIUSD:

• Being a stablecoin, maintained a price close to 1.0000 USDT
• Occasional spikes above 1.0025 USDT indicate temporary deviations from the peg
• Increased volatility was observed in early February 2025, possibly due to liquidity shifts or market events
• Overall, the price remained within an expected range, reinforcing its stability

Timeseries for EUR as Comparison

Distribution of Close for MKRUSDT

Key observations:

• Bimodal distribution with main peak at 1500-1600 USDT
• Secondary peak around 1200 USDT, right-skewed with tail extending to 2400
• Majority of price action concentrated between 1400-1800 USDT range

Machine Learning Models

Target Variable Analysis: y

Mean y: 1562.8944
Standarddeviation y: 62.9891

Key observation:

• Illustrates the frequency distribution of the target variable y across the Train, Validation, and Test datasets
• Range: Values spread from ~1000 to 2400
• Main concentration: Highest density around 1400-1600
• Shape: Right-skewed distribution (longer tail towards higher values)
• Dataset splits show similar patterns across train/validation/test

Key observations:

• Training Set: widest spread of values, shows several outliers (dots above box), median around 1500, larger variability compared to other sets
• Validation Set: more compact distribution, fewer outliers, similar median to training set, more consistent/stable values
• Test Set: lower median compared to train/validation, smaller spread, few outliers, suggests slightly different market conditions

Linear Regression (LR)

Mean Squared Error (MSE) on reduced Validation Set: 173.01
Mean Absolute Error (MAE) on reduced Validation Set: 8.57
R-squared (R²) on reduced Validation Set: 0.96

• Fibonacci_0 has the most significant impact on model accuracy with an accuracy drop of nearly -10 when removed
• Other important features include intraday_range and ln_volume
• Most Fibonacci-based features (23.6, 38.2, 50, 100) show minimal impact, suggesting they could be removed to simplify the model without significant performance loss
• Technical indicators like RSI, MACD have surprisingly low importance

Key Observations:

• Distribution of predictions spans 1450-1700, with most values between 1500-1650
• Multi-modal distribution suggests distinct price levels or regimes in the data
• Relatively few predictions at extreme values (1450 or 1700) indicates the model is conservative in predicting extreme prices
• Irregular shape suggests the underlying price movements aren't normally distributed, which is typical for financial data

Decision Trees (DT)

Best Hyperparameters: max_depth 10.00000
min_samples_leaf 24.00000
mse 205.15055
Decision Tree Regressor MSE on the validation set: 205.151
Decision Tree Regressor R2 Score on the validation set: 0.948

Key observations:

• Strong inverse relationship between max_depth and MSE
• MSE decreases significantly as max_depth increases from 1.0 to 4.0
• Different min_samples_leaf values show very similar performance curves
• Deeper trees (max_depth=4.0) provide best predictive performance
• Further increasing max_depth likely risks overfitting
• Simple configuration with min_samples_leaf=1 sufficient for good results

Random Forest (RF)

Validation MSE: 194.431
RMSE: 13.944
R² Score: 0.951

Key observations:

• min_samples_leaf=10 shows best performance (lowest MSE ~195)
• Higher min_samples_leaf values provide more stable learning curves
• Smaller leaf sizes (1-3) show higher variance and worse performance
• Clear separation between different leaf sizes indicates this is a crucial parameter

Key observations:

• Points roughly symmetrically scattered around zero line (red)
• No clear funnel or megaphone pattern suggests homoscedasticity
• Most residuals appear within reasonable bounds (-2 to +2)
• Consistent variance across all predicted values (1450-1700), with few outliers present, particularly in higher prediction ranges
• No systematic bias visible (even distribution above/below zero)
• Slight heteroscedasticity at extreme values (1450 and 1700)
• Some notable outliers around 1500 and 1650 prediction ranges
• Denser point concentration in 1500-1600 range suggests more training data here

XGBoost

Best Hyperparameters: eta 0.150000
max_depth 4.000000
min_child_weight 10.000000
rmse 13.602891

XGBoost MSE on the Validation Set: 185.0386
XGBoost MAE on the Validation Set: 9.2880
XGBoost R² Score on the Validation Set: 0.9534

Key observations:

• Very strong linear relationship (R² likely >0.95)
• Points cluster tightly around ideal prediction line
• Slight heteroscedasticity - spread increases at higher values
• Few outliers, mostly in upper value range (>1650), more uncertainty in predictions above 1650
• Model maintains accuracy across entire value range
• No systematic bias (points evenly distributed above/below line)
• Slight tendency to underestimate extreme high values
• Very reliable predictions in middle range (1500-1600)
• No evidence of overfitting or underfitting

Key observations:

• Most influential features are Fibonacci levels, especially fibonacci_0, fibonacci_23_6, and fibonacci_50, which dominate the model’s decisions
• Traditional indicators like moving averages (sma10, sma20) and volume-based indicators have minimal impact, suggesting they may not add much predictive power
• This insight could be used to refine feature selection, possibly reducing dimensionality and improving model efficiency

Key observations:

• fibonacci_0 dominates with highest impact (~150 magnitude)
• Fibonacci retracement levels are strongest predictors
• Secondary group (fibonacci_23_6, fibonacci_50, sma10) shows moderate impact (20-40)
• Simple moving averages outperform complex indicators
• Many common technical indicators have negligible predictive power
• Most technical indicators (MACD, ATR, CCI etc.) show minimal impact (<5)
• Model heavily relies on basic price action vs technical analysis

The best model is:

XGBoost MSE on the Validation Set: 185.0386
XGBoost MAE on the Validation Set: 9.2880
XGBoost R² Score on the Validation Set: 0.9534

Best Hyperparameters:
eta 0.150000
max_depth 4.000000
min_child_weight 10.000000
rmse 13.602891

Installation

• Clone the repository:

  git clone https://github.com/your-repo.git
  cd your-repo

• Set up the environment Using Conda:

  conda env create -f environment.yml
  conda activate your-environment-name

• Using pip:

  pip install -r requirements.txt

Installation Instructions for TA-Lib

• TA-Lib library is required for this project but is not installed automatically via the environment.yaml file
• You need to install it manually due to potential platform-specific compilation requirements

To install TA-Lib, follow these steps:

• Using Conda (Recommended):

conda install -c conda-forge ta-lib

• Using pip: If you prefer pip, ensure you have the required dependencies installed and run:

pip install TA-Lib

• On macOS with Homebrew: First, install the TA-Lib C library:

brew install ta-lib

• Then install the Python wrapper:

pip install TA-Lib

• On Linux: Install the required development library (e.g., for Ubuntu):

sudo apt-get install libta-lib-dev

• Then install the Python wrapper:

pip install TA-Lib

• On Windows: Download and install the precompiled binaries for your system from the TA-Lib website, then install the Python wrapper:

pip install TA-Lib

Make sure TA-Lib is installed before running the application. If you encounter any issues, refer to the TA-Lib documentation for further assistance

How to Use

• Jupyter Notebook (notebook.ipynb):

  • Fetch cryptocurrency data:
    • Fetches data for cryptocurrencies and stablecoins defined in the coins list
    • Logs missing or delisted cryptos as warnings or errors
    • Processes the data and adds derived metrics (e.g., price change)
    • Saves the final dataset as one_df.csv
  • Perform feature engineering and derive metrics
  • Evaluate multiple machine learning models
  • Save the best models as .pkl files

• Train the Model:

  • Use train.py to train the best-performing model (default: XGBoost) on the processed data
  • Save the trained model as a .pkl file

• Deploy the Model with Flask:

  • Use predict.py to deploy the model and provide predictions via Flask

Flask

• The repository includes a Flask (predict.py) to interact with the trained XGBoost model. The API allows users to predict the 'close' within the next hour.

• Steps to Use:

  • Start the Flask Server
  • Ensure the conda environment is active and run:

    python predict.py --port=<PORT>

• Replace with the desired port number (e.g., 5001). If no port is specified, the server defaults to port 8000

• Example:

    python predict.py --port=5001

• The server runs at http://0.0.0.0:, for example: http://0.0.0.0:5001

Make Predictions

• Send an HTTP POST request with the input features as JSON to the /predict endpoint. Replace with the port you specified earlier

• Example Input:

curl -X POST http://127.0.0.1:5001/predict \
-H "Content-Type: application/json" \
-d '{"ln_volume": 8.573291254393968, "bop": -0.75, "ppo": 1.6623794774736873}'

• Example Response:

{
"predicted_close_price": 1098.2784423828125
}

Run with Docker

• To simplify deployment, a Dockerfile is provided. To build and run the Docker container:

• Build the Docker image:

docker build -t mkr-coin-analysis .

• Run the container:

docker run -p 5001:5001 mkr-coin-analysis

License

This project is open-source and licensed under the MIT License.