Loading repository data…
credough / repository
End-to-end Bitcoin price analytics pipeline. Ingests live market data via CoinGecko API, stores records in PostgreSQL, computes moving averages, volatility, and buy/sell signals, and visualizes insights through an interactive Streamlit dashboard. Built in Python.
A real-time data pipeline that ingests, stores, and analyzes Bitcoin price data — surfacing actionable market insights through an interactive analytics dashboard.
Cryptocurrency markets move fast and generate large volumes of price data that are difficult to interpret without structure. Raw price feeds alone offer little analytical value; the challenge is transforming continuous market data into meaningful signals that support decision-making.
This project addresses that challenge by building an end-to-end pipeline that:
The goal is not to predict the market with certainty, but to provide a clear, data-driven view of Bitcoin's recent behavior — the kind of foundation any analyst or decision-maker would need before drawing conclusions.
Source: CoinGecko Public API (no authentication required)
Endpoints used:
/simple/price — real-time price, market cap, 24h volume, and 24h percentage change/coins/bitcoin/market_chart — historical daily price, market cap, and volume (up to 90 days)What the data contains:
| Field | Description |
|---|---|
timestamp | Date and time of the record |
price_usd | Bitcoin price in US dollars |
market_cap | Total market capitalization |
volume_24h | Total trading volume over the preceding 24 hours |
change_24h | Percentage price change over the preceding 24 hours |
Limitations and challenges:
change_24h field is not available in the historical endpoint and is only captured during live fetches, resulting in null values for seeded recordsTimestamp normalization
The historical endpoint returns timestamps in Unix milliseconds. These are converted to timezone-aware datetime objects (TIMESTAMPTZ) during ingestion to ensure consistent querying and correct time-series alignment across live and historical records.
Hourly aggregation for analytics
Live fetches can produce multiple records per hour depending on scheduler frequency. Before analytics are computed, records are grouped and averaged by hour using a SQL aggregation query. This prevents duplicate data points from inflating moving averages and distorting signal detection.
Duplicate prevention
A row count check runs before any historical seed operation. If the database already contains more than 10 records, seeding is skipped. This protects against accidental data duplication when the seed script is re-run.
Numeric precision
SQLite's REAL type was replaced with PostgreSQL's NUMERIC(18,2) and NUMERIC(24,2) for price and volume fields respectively. Floating-point imprecision in financial data compounds over time, making precise decimal storage essential for reliable aggregations.
The dataset covers 90 days of daily Bitcoin price history supplemented by live fetches. Several patterns emerged during exploration:
Price behavior is non-linear but trend-following over medium windows. Short-term noise is significant, but when smoothed with a 7-day or 30-day moving average, clear directional trends become visible. The gap between these two averages provides a reliable indicator of momentum shifts.
Volatility is cyclical, not random. Periods of high volatility consistently precede or follow major price movements. Observing volatility trends — rather than price alone — provides earlier warning of potential market shifts.
Buy and sell signals are sparse but meaningful. MA crossover events (where the 7-day average crosses the 30-day average) are infrequent, which makes each signal relatively significant. A high ratio of BUY to SELL signals over a 30-day window correlates with sustained upward price movement.
24-hour change is a noisy but useful sentiment indicator. Daily percentage changes fluctuate widely, but sustained positive or negative streaks in this metric often reflect broader market sentiment rather than isolated price events.
What it shows: A real-time view of the four most important market metrics: current price, 24-hour change, total records in the database, and the all-time high within the dataset.
Key takeaway: Provides an at-a-glance summary of current market conditions and confirms that the data pipeline is actively collecting data.
What it shows: A time-series line chart of Bitcoin's USD price across all stored records, with a filled area beneath the line to emphasize directional movement.
Key takeaway: The overall price trajectory and key inflection points are immediately visible, giving context to any single data point.
What it shows: A bar chart of daily trading volume, reflecting market participation and liquidity over time.
Key takeaway: Volume spikes frequently coincide with significant price movements, reinforcing the connection between market activity and price action.
What it shows: A bar chart of daily percentage price change, with positive and negative values color-coded to reflect bullish and bearish days.
Key takeaway: Streaks of consecutive positive or negative days are more informative than any single day's change, and are clearly visible in this view.
What it shows: Bitcoin's price overlaid with 7-day and 30-day moving averages. Crossover points between the two averages are the basis for the buy/sell signal logic.
Key takeaway: The relationship between the 7d and 30d MA tells a cleaner story than raw price — it filters out daily noise and highlights sustained momentum shifts.
What it shows: Price chart annotated with buy (triangle-up) and sell (triangle-down) signals generated by MA crossover detection. Each signal marks a point where the short-term average crossed the long-term average.
Key takeaway: Signals are deliberately infrequent, which increases their reliability. The ratio of buy to sell signals over any given window is a practical summary of trend direction.
What it shows: Rolling 7-day standard deviation of Bitcoin's price, displayed as a filled area chart. Higher values indicate greater price instability.
Key takeaway: Volatility is not constant — it compresses and expands in cycles. Identifying when volatility is rising early allows for more informed risk assessment before a price move fully develops.
What it shows: A linear regression model trained on the selected historical window, extended 7 days into the future. The actual price and the forecast line are overlaid for direct comparison.
Key takeaway: The forecast captures the dominant trend direction, which is useful for understanding whether recent price action is consistent with or diverging from the broader trend.
Data moves through four stages from source to visualization:
CoinGecko API
|
v
Python Ingestion Layer (pipeline.py)
- Fetches live price data on a scheduled interval
- Validates and structures the API response
|
v
PostgreSQL Database
- Stores all records with timezone-aware timestamps
- Supports efficient time-range queries for analytics
|
v
Analytics Engine (analytics.py)
- Computes moving averages, volatility, signals, and forecast
- Aggregates data by hour to ensure clean inputs
|
v
Streamlit Dashboard (dashboard.py)
- Renders live and analytics views in a two-tab interface
- All metrics and charts update dynamically based on user controls
| Layer | Tool |
|---|---|
| Data Source | CoinGecko REST API |
| Ingestion | Python, Requests |
| Storage | PostgreSQL, SQLAlchemy, psycopg2 |
| Analytics | Pandas, NumPy, scikit-learn |
| Visualization | Streamlit, Plotly |
| Scheduling | schedule (Python library) |
| Environment | python-dotenv |
| Version Control | Git |
Aaron Creed P. Celindro
Selected from shared topics, language and repository description—not editorial ratings.
yannpointud /
End-to-end crypto price direction prediction framework built on the Mamba Selective State Space architecture (Python/PyTorch). Covers the full lifecycle: data preparation, Triple Barrier labeling, training, evaluation & live inference — all from a single config. Linear-time sequence processing as a Transformer alternative.
priyadharshinivetrivel /
End-to-End Bitcoin Price Analytics and Forecasting using XGBoost and Support Vector Regression
trivenigathibandhe-coder /
End-to-end Bitcoin price analysis and 90-day forecasting using Python, yfinance, Prophet, and time-series analysis.
Hvllvix /
This project implements an end-to-end machine learning workflow to predict Bitcoin's closing price.
Ashkimo0097 /
An end-to-end time-series project that analyzes historical Bitcoin price data, engineers technical indicators, predicts next-day price direction using ML classifiers, and forecasts future prices using LSTM with time-series cross-validation.
Analyzed cryptocurrency market trends using Python for data preprocessing, SQL for calculations (CTEs, window functions, joins), and Power BI for visualizations (rolling averages, volatility tracking). Conducted price prediction, risk analysis, correlation studies, and anomaly detection across Bitcoin, Ethereum, Litecoin, and Tether datasets.