Financial Time-Series Anomaly Detection

Project Overview

This project focuses on detecting anomalies in financial time-series data, specifically stock price trends. The primary goal is to identify unusual behavior or outliers in stock movement that could signal major events, fraud, or trading opportunities.

We use a combination of statistical techniques and machine learning algorithms to analyze historical stock price data and flag anomalies in trends, volatility, and price fluctuations.

---

Objectives

• Load and preprocess historical stock market data
• Calculate key financial indicators (moving averages, volatility, returns)
• Apply anomaly detection algorithms (Isolation Forest, DBSCAN)
• Forecast trends using LSTM or Prophet
• Visualize anomalies and prediction uncertainty

---

Dataset

• Source: Yahoo Finance API
• Sample Ticker: AAPL, MSFT, TSLA, etc.
• Time Period: Adjustable (e.g., last 5 years)
• Features Used:
  • Open, High, Low, Close, Volume
  • Technical indicators (Moving Average, RSI, Volatility, etc.)

You can change the ticker and time range to analyze different stocks.

---

Technologies Used

• Python
  • pandas, numpy – data manipulation
  • matplotlib, seaborn, plotly – visualization
  • scikit-learn – Isolation Forest, DBSCAN
  • tensorflow / keras – LSTM (forecasting)
  • prophet – Facebook Prophet time-series model
• yfinance – fetch real-time and historical stock data

---

Project Workflow

1. Data Collection

   • Fetch historical stock data using yfinance

2. Preprocessing

   • Handle missing values, compute log returns
   • Calculate rolling averages and volatility

3. Anomaly Detection

   • Apply Isolation Forest and/or DBSCAN on engineered features
   • Label points as anomalous or normal

4. Forecasting

   • Use LSTM or Prophet to model and forecast future stock trends
   • Evaluate using RMSE, MAE

5. Visualization

   • Plot stock prices with anomalies highlighted
   • Forecast with uncertainty intervals

---

Results & Insights

• Successfully detected price spikes/drops using Isolation Forest
• DBSCAN detected density-based outliers in low-volume periods
• Prophet forecasted trends with seasonality and trend changepoints
• LSTM showed potential for multi-step forecasting but required tuning

Anomalies can help in building alert systems for financial risk or trading signals.

---

How to Run

1. Clone the repository:

   git clone https://github.com/yourusername/financial-anomaly-detection.git
   cd financial-anomaly-detection