Bitcoin Price Prediction using ARIMA

This project demonstrates end-to-end time series forecasting of Bitcoin prices using ARIMA and SARIMA models in Python.
The notebook provides a complete workflow: data cleaning, feature engineering, decomposition, model selection, prediction, interactive visualization, and accuracy evaluation.

---

Project Overview

The goal is to model and forecast Bitcoin's monthly price trends by leveraging statistical time series techniques.
Key highlights include:

• Handling real-world minute-level price data
• Data resampling and creation of new features
• Trend and seasonality decomposition
• Stationarity and transformation
• Model parameter tuning and grid search
• Forecast visualization with Plotly

---

Dataset

• Source: Bitcoin Historical Data (Kaggle)
• File: btcusd_1-min_data.csv
• Description:
  This dataset contains minute-by-minute OHLCV (Open, High, Low, Close, Volume) data for Bitcoin, starting from January 1, 2012.
  It provides over 13.5 years of continuous price and volume data, making it one of the most detailed long-term Bitcoin datasets available for model training and evaluation.

---

Workflow & Methodology

1. Data Loading & Preparation

• Load raw CSV data.
• Convert Unix timestamps to readable datetime.
• Set timestamp as index for resampling.

2. Feature Engineering

• Calculate Weighted Price: average of High, Low, and Close.
• Compute VWAP (Volume Weighted Average Price) for daily, monthly, quarterly, and yearly periods.

3. Visualization

• Use Plotly to visualize price trends over different timeframes.
• Create interactive multi-panel charts for daily, monthly, quarterly, and yearly data.

4. Trend & Seasonality Analysis

• Apply STL decomposition to split price series into trend, seasonality, and residuals.

5. Data Transformation & Stationarity

• Apply Box–Cox transformation to stabilize variance.
• Use differencing (seasonal and regular) to achieve stationarity, validated with Dickey-Fuller tests.

6. Autocorrelation Analysis

• Plot ACF and PACF to inform ARIMA/SARIMA parameter selection.

7. Model Selection & Fitting

• Grid search over possible ARIMA/SARIMA parameter combinations.
• Select the best model based on AIC (Akaike Information Criterion).

8. Model Diagnostics

• Analyze model residuals to ensure assumptions are met.

9. Forecasting

• Generate predictions for future months (e.g., up to January 2026).
• Apply inverse Box–Cox transformation to revert predicted values to their original scale.
• Visualize actual vs. predicted prices.

10. Model Evaluation

• Split data into training and testing periods.
• Compute forecast accuracy using MAPE (Mean Absolute Percentage Error).

---

Live Prediction

You can view the live prediction on kaggle here https://www.kaggle.com/code/dairus/bitcoin-price-prediction-by-time-series-and-arima/notebook

Usage

1. Clone this repository and open the notebook in Jupyter:

   git clone https://github.com/Dairus01/Bitcoin-Price-Prediction
   cd Bitcoin-Price-Prediction
   jupyter notebook Bitcoin Price Prediction.ipynb