**Authors:** Dr. B. Chandrashekar, C. Yosepu, Dr. Saikrishna Boggavarapu, Dr. Sanjeev Chepyala, Sumit Pundir, Dr. G. Manikandan
**Published:** _6th International Conference on Contemporary Computing and Informatics (IC3I)_, 2023
**Link:** [IEEE Xplore](https://ieeexplore.ieee.org/)
---
### Summary
This study proposes a hybrid financial forecasting model combining #ARIMA and #Gradient-Boosting to improve the accuracy of stock price predictions in the stock exchange market. It leverages ARIMA for time-series analysis and Gradient Boosting for non-linear pattern learning to achieve enhanced forecasting performance.
---
### Key Points
- **ARIMA for Time-Series Forecasting:**
- Captures #temporal dependencies and long-term #trends.
- Effective for both #stationary and #non-stationary financial data.
- **Gradient Boosting for Non-Linear Patterns:**
- Learns complex relationships missed by traditional statistical models.
- Uses residuals from ARIMA as input to refine predictions.
- **Hybrid Approach:**
- ARIMA is first applied to pre-process the time series and extract residuals.
- Gradient Boosting then learns the non-linear features from these residuals.
- This combination improves the model’s ability to **predict sudden changes** in stock prices.
---
### How It Works
1. **Data Collection:** Historical financial data including stock prices, trading volumes, and indices.
2. **Data Pre-processing:** Handles missing values, outliers, and normalizes data for ARIMA modeling.
3. **ARIMA Modeling:** Analyzes temporal relationships and generates residuals.
4. **Feature Engineering:** Residuals from ARIMA are treated as new features.
5. **Gradient Boosting Modeling:** Learns non-linear relationships in the data.
6. **Model Training and Evaluation:** Uses Mean Absolute Error #MAE, Mean Squared Error #MSE, and #R² for performance assessment.
7. **Deployment and Monitoring:** Deployed for real-time financial forecasting with ongoing monitoring.
---
### Performance Results
- Achieved **97.12% accuracy**, outperforming #LSTM (90.54%), SVM (92%), and Linear Regression (88.12%).
- **Mean Absolute Error:** 105.361 (lowest among all compared models)
- **Mean Squared Error:** 57.547 (significantly lower than other methods)
- **R² Score:** 0.993, indicating very high predictive reliability.
---
### Strengths
- Combines both #temporal analysis and #non-linear learning.
- Outperforms conventional models in terms of prediction accuracy.
- Adapts well to **volatile market conditions** with more precision.
---
### Limitations
- Performance may vary with **different market conditions**.
- Relies heavily on high-quality historical data for accuracy.
- Optimization of hyperparameters is crucial for best results.