ADS Capstone Chronicles Revised

13

Data preparation included creating new features and normalizing data for accurate analysis and modeling. Correlation heatmaps identified multicollinearity, helping refine the dataset for better model performance. Distribution analysis of percentage changes across various cryptocurrencies highlighted their volatility, while average percentage change by day of the week revealed significant mid-week peaks. Stationarity tests confirmed the reliability of historical price data, laying a robust foundation for predictive modeling, which is crucial for empowering users with accurate investment decisions on the platform. 5 Feature Engineering Feature engineering is a practical and effective tool in preparing data for time series modeling and forecasting. By creating new features, the underlying patterns and trends in the data can be captured more effectively, thus improving the accuracy of predictive models. 5.0.1 Percentage Change One of the primary features added is the percentage change column ‘ pct_change ’ for the closing price. This column is crucial as it measures the price change over a specific period and is intended to be used as the target variable in modeling and forecasting. 5.0.2 Lag Features Several lag features for columns, such as close and volume, were added over various time windows (7, 14, and 30 days). These features help identify trends and patterns over time in the raw data, such as short-term and long-term stock price trends or trading volume trends. 5.0.3 Rolling Statistics Rolling statistics were added to the dataset for multiple windows, including rolling mean and standard deviation. These statistics smooth out short-term fluctuations and highlight longer-term trends. Exponential Moving Averages (EMA)

were also calculated, providing a weighted average of past data that decreases exponentially.

5.0.4 Bollinger Bands Bollinger Bands, consisting of an upper band, a lower band, and a moving average, were created using a 20-day window. These bands help identify overbought or oversold conditions in the market by measuring volatility and providing a dynamic range for price movement. 5.0.5 Momentum Indicators The stochastic oscillator and Williams %R are momentum indicators that compare a particular closing price to a range of similar prices over a certain period. By analyzing the speed and momentum of price movements, these indicators help identify potential reversal points. 5.0.6 Volume Oscillator A volume oscillator was added to capture the difference between short-term and long-term volume trends. This indicator helps identify periods of significant trading activity, such as sudden spikes in trading volume, which often precede substantial price changes. 5.0.7 Other Features Other features were engineered to capture different aspects of the data, including the month, quarter, high-low ratio, close-open ratio, and logarithmic returns. These features provide a comprehensive view of the market dynamics, enabling more accurate predictions. Values The feature engineering process introduced new null and zero values in the dataset. These were handled appropriately by applying forward, and backward fill methods and mean imputations were used for numerical columns. Forward and backward fill ensure the continuity of data, which is crucial for time series analysis, while mean imputation for numerical columns maintains the dataset's integrity. 5.0.8 Handling New Null and Zero

80