📈 Time Series Forecasting – Stock Prices

📂 Project Overview

1. Dataset Acquisition
   • Uses the Kaggle dataset: World Stock Prices – Daily Updating
   • Downloaded via kagglehub
2. Preprocessing
   • Convert Date column to datetime and set as index.
   • Sort data by date.
   • Filter for a single stock (default: AAPL).
   • Handle missing values using forward-fill / interpolation.
3. Exploratory Data Analysis (EDA)
   • Visualize stock closing prices over time.
   • Check for stationarity (ADF test).
   • Plot ACF/PACF to guide ARIMA orders.
4. Modeling Approaches
   • ARIMA: Baseline forecasting with manually selected parameters.
   • Auto-ARIMA: Automated parameter tuning using AIC/BIC.
   • SARIMAX: Seasonal ARIMA with exogenous variables (optional).
   • LSTM: Deep learning model for sequence prediction.
5. Forecast Evaluation
   • Implement rolling window forecasting:
     • Train on initial window → Forecast one step ahead → Update with true value → Repeat.
   • Evaluate models using metrics:
     • RMSE, MAE, MAPE

⚙️ Installation & Requirements

🚀 Usage

1. Clone or download the notebook:
   bash

   1git clone <repo-link> 2cd stock-forecasting
2. Run the Jupyter Notebook:
   jupyter notebook "Time_Series_Forecasting_–_Stock_Prices.ipynb"
3. Inside the notebook:
   • Change the stock ticker (ticker_symbol = 'AAPL') to forecast other stocks.
   • Choose between ARIMA / SARIMAX / LSTM sections.
   • Run evaluation cells to compare results.

📊 Results

• ARIMA works well for short-term stationary series.
• SARIMAX captures seasonality but is computationally heavier.
• LSTM adapts to nonlinear patterns, showing better performance for long-term forecasts.
• Rolling window evaluation provides realistic performance assessment compared to a single train-test split.

📌 Key Files

• Time_Series_Forecasting_–_Stock_Prices.ipynb → Main notebook with full workflow
• world_stock_prices_2_years.csv → Example filtered dataset for a single stock

✨ Future Improvements

• Add Prophet model for easier seasonality handling.
• Incorporate financial indicators (RSI, MACD) as exogenous features.
• Deploy as a web dashboard with live forecasts.

📜 License