₿ Smart Crypto Price Predictor

An end-to-end Machine Learning project that predicts Bitcoin (BTC-USD) closing prices using historical market data, technical indicators, and an interactive Streamlit dashboard.

---

Overview

This project builds a complete ML pipeline for cryptocurrency price prediction — from raw data ingestion and feature engineering to model training, evaluation, and a live interactive UI. The goal is to predict the next Bitcoin closing price using historical OHLCV data combined with technical indicators commonly used in financial analysis. Two models are compared (Linear Regression and Random Forest), with full evaluation metrics and a 7-day iterative forecast.

---

Live Demo :

---

Video Demo :

---

Features

Price Chart Tab

• Interactive candlestick chart with MA7 and MA30 overlay
• Dedicated RSI (14-day) panel with overbought/oversold zones highlighted
• Volume bars color-coded green/red based on candle direction
• Adjustable slider to view last N days of history
• Download CSV button for filtered data

Predict Tab

• Auto-filled inputs from latest market data
• Predicts next Bitcoin closing price using trained model
• Confidence range shown as Predicted Price ± RMSE
• Bullish / Bearish / Neutral signal badge based on predicted movement

7-Day Forecast Tab

• Iterative 7-day forecast — each day's prediction feeds into the next
• Forecast cards with day labels and % change from current price
• Forecast chart with confidence band overlay

Model Performance Tab

• Evaluation metrics: MAE, RMSE, MAPE
• Actual vs Predicted line chart (test set)
• Feature importance horizontal bar chart
• Residuals bar chart (green/red)

Sidebar Controls

• Toggle between CSV file and Live yfinance data
• Switch between Linear Regression and Random Forest models
• Last updated timestamp displayed in header

---

ML Pipeline

Raw CSV / yfinance → Preprocessing → Feature Engineering → Train/Test Split (shuffle=False) → Model Training → Evaluation → Streamlit UI

---

Feature Engineering

Feature	Description
Open, High, Low	Daily OHLC price values
Volume	Daily trading volume
MA7	7-day moving average
MA30	30-day moving average
Lag1, Lag2, Lag3	Previous 1, 2, 3 day closing prices
Returns	Daily percentage return (pct_change)
RSI	Relative Strength Index (14-day)

---

Tech Stack

Category	Tools
Language	Python 3.9+
Data Handling	Pandas, NumPy
Live Data	yfinance
Modeling	Scikit-learn (LinearRegression, RandomForestRegressor)
Visualization	Matplotlib
Frontend	Streamlit
Serialization	Joblib

---

Project Structure

smart-crypto-price-predictor/
│
├── data/
│   └── processed_btc_data.csv      # Preprocessed BTC-USD dataset
│
├── notebooks/
│   ├── eda.ipynb                   # Exploratory Data Analysis
│   ├── features.ipynb              # Feature engineering walkthrough
│   └── model.ipynb                 # Model training & evaluation
│
├── src/
│   ├── preprocess.py               # Data cleaning & feature generation
│   ├── train.py                    # Model training script
│   ├── predict.py                  # Prediction logic
│   ├── data_loader.py              # Data loading utilities
│   └── utils.py                    # Helper functions
│
├── models/
│   └── model.pkl                   # Saved trained model
│
├── app/
│   └── app.py                      # Streamlit frontend
│
├── requirements.txt
└── README.md

---

Getting Started

# 1. Clone the repo
git clone https://github.com/hassan-ali786/smart-crypto-price-predictor.git
cd smart-crypto-price-predictor

# 2. Install dependencies
pip install -r requirements.txt

# 3. Run the app
streamlit run app/app.py

If models/model.pkl is not present, the app will auto-train from the CSV on first launch.

---

Results

Model	MAE	RMSE	MAPE
Linear Regression	~$1,200	~$1,800	~3.2%
Random Forest	~$900	~$1,400	~2.4%

Results on 20% holdout test set (time-series split, shuffle=False)

---

Authors

This project was developed collaboratively by:

• Hassan Ali
• Dua Zahra

Both contributors worked together on data analysis, machine learning development, implementation, and project integration.

---

Connect