AAPL Stock Price Forecasting

Forecasting Apple Inc. (AAPL) closing stock prices using ARIMA, SARIMA, and LSTM models.
This project compares traditional statistical forecasting with deep learning to understand how each performs on real-world financial time series data.

---

Project Summary

Stock forecasting is a key challenge in financial data science.
Here, historical AAPL data (2018–2025) is used to predict future closing prices.
The notebook walks through:

• Building ARIMA/SARIMA models for trend-based forecasting
• Training an LSTM network for sequence learning
• Comparing performance and forecast accuracy

---

Repository Structure

AAPL_Stock_Price_Forecasting/
│
├── Data/
│ └── AAPL_historical.csv
│
├── Notebooks/
│ └── stock_forecasting_pipeline.ipynb
│
└── README.md

---

Approach

1. Data Preparation

   • Pulled daily AAPL data using yfinance
   • Cleaned missing values and ensured business-day frequency

2. ARIMA & SARIMA

   • Performed grid search for optimal (p, d, q) parameters
   • Selected ARIMA(2, 2, 3) (lowest AIC = 6165.47)
   • Tested SARIMA for yearly seasonality — smoother but similar accuracy

3. LSTM

   • Scaled data with MinMaxScaler
   • Used 60-day lookback windows
   • Built a multi-layer LSTM model with dropout and early stopping
   • Produced sharper, short-term accurate forecasts

4. Evaluation

   Model	MAE	RMSE	Observation
   ARIMA(2, 2, 3)	14.6883	17.9199	Stable but lags during sharp changes
   LSTM	6.2073	7.7119	Tracks volatility and adapts faster

---

Key Insights

• LSTM outperformed ARIMA/SARIMA with ~2.5× lower RMSE.
• ARIMA/SARIMA captured long-term trends but missed short-term swings.
• Both flatten when forecasting beyond 6 months — highlighting uncertainty.
• A Hybrid ARIMA–LSTM could combine interpretability and adaptability.

---

Tools & Libraries

Pandas • NumPy • Statsmodels • TensorFlow/Keras • Scikit-learn • Matplotlib

---

Future Work

• Implement a Hybrid ARIMA–LSTM model
• Add market indicators like trading volume or sentiment
• Explore Transformer-based time series architectures
• Build a Streamlit dashboard for visualization

---

Author

Laith Waqas Mohammed
Data Science Student | Financial Forecasting Enthusiast
Dublin, Ireland
LinkedIn