ICRCET Yelloppoya University Conference Presentation
- 1. Enhanced LSTM-NeuralProphet Model for Share Price Forecasting Siva A,Sedhu S and ThillaiGovinda P Under the Guidance of J. Swarnalakshmi Presented by Siva 14th International Conference on Recent Challenges In Engineering And Technology (ICRCET-2025)
- 2. Biography:- 14th International Conference on Recent Challenges In Engineering And Technology (ICRCET-2025) I’m Siva A, a final-year B.Tech CSE student at Hindustan Institute of Technology and Science, with strong skills in Python and Java. Along with my teammates Sedhu and Thillai, we built the project “Enhanced LSTM-NeuralProphet Model for Share Price Forecasting” under the guidance of Mrs. Swarnalaskhmi J. •Sedhu contributed to the LSTM architecture. •Thillai handled deployment and results analysis. Our goal was to create an accurate and scalable stock forecasting model.
- 3. Abstract:- 14th International Conference on Recent Challenges In Engineering And Technology (ICRCET-2025) We propose an improved LSTM-NeuralProphet hybrid model for Indian stock price prediction. It combines multi-scale LSTM to capture long-term dynamics and NeuralProphet for trends, seasonality, and external factors. The model adapts to market shifts and demonstrates higher accuracy using metrics like MAPE, RMSE, and R². Feature selection techniques such as PCA and correlation-based methods enhance input quality.
- 4. Description:- 14th International Conference on Recent Challenges In Engineering And Technology (ICRCET-2025) Goal: To build a highly accurate model for predicting Indian stock prices. Methodology: •Combined LSTM (for long-term patterns) with NeuralProphet (for trends and seasonality). •Trained on NIFTY 50 and major Indian stocks using Yahoo Finance data. Tools: Python, LSTM, NeuralProphet, Yahoo Finance API Result: •Predictions closely match real-time closing prices. •Outperformed traditional models in accuracy and error metrics. •Scalable for crypto and forex forecasting.
- 5. Architecture Diagram:- 14th International Conference on Recent Challenges In Engineering And Technology (ICRCET-2025)
- 6. Proposed Method:- 14th International Conference on Recent Challenges In Engineering And Technology (ICRCET-2025) 1. Data Collection & Preprocessing: Acquire historical stock price data. Handle missing values and perform feature engineering. 2. Data Splitting: Divide the dataset into training (80%) and testing (20%). 3. Model Training: Train LSTM Model on time-series data.Train NeuralProphet Model separately for additional forecasting insights. 4. Model Evaluation: Compare performance using RMSE, MAE, and R².Validate results using unseen data. 5. Final Prediction & Explainability: Combine results for improved accuracy. Use NeuralProphet to analyse trends, seasonality, and outliers.
- 7. Conclusion: 14th International Conference on Recent Challenges In Engineering And Technology (ICRCET-2025) •Built a highly accurate LSTM model through multiple training and testing cycles with optimized (fewer) epochs. •Developed a NeuralProphet model that delivered enhanced forecasting accuracy. •Predicted share prices closely matched real-time closing values, demonstrating strong model reliabilit
- 8. Evaluation Metrics: 14th International Conference on Recent Challenges In Engineering And Technology (ICRCET-2025) NSE ID R-Square Accuracy% MAPE % RELIANCE 0.35 96.48 3.52 TCS 0.6502 97.37 2.63 HCL 0.81 96.74 3.26 WIPRO 0.91 96.81 3.91 INFY 0.42 88 3.2 The table above highlights the values of the accuracy metrics of LSTM NSE ID R-Square RMSE% MAPE % RELIANCE 0.712 12 0.82 TCS 0.65 38.85 0.92 HCL 0.323 88 4.1 WIPRO 0.194 4.69 1.86 HDFCBANK 0.5959 29.54 1.4 The table above highlights the values of the accuracy metrics of NeuralProphet
- 9. Result: 14th International Conference on Recent Challenges In Engineering And Technology (ICRCET-2025) Visualized output of LSTM with metrics rate are R² Score: 0.6495,MAE: 84.8706,RMSE: 103.2111,MAPE: 2.64%,Accuracy: 97.36% Visualized output of Neuralprophet with metrics rate are R² Score :0.6716, RMSE: 37.37,MAPE: 0.98%
- 10. Result: 14th International Conference on Recent Challenges In Engineering And Technology (ICRCET-2025) Real-Time Close price is 1443.85 Visualized output of Neuralprophet for Infosys Predicted price around 1475 to 1438
- 11. Future Works: 14th International Conference on Recent Challenges In Engineering And Technology (ICRCET-2025) •Deploy the model in Streamlit with support for parallel predictions of multiple stocks. •Extend the forecasting system to work with cryptocurrencies like Bitcoin and Ethereum. •Integrate real-time news sentiment analysis to enhance prediction accuracy.
- 12. Appendix: 14th International Conference on Recent Challenges In Engineering And Technology (ICRCET-2025) •Dataset Info: • Source: Yahoo Finance API •Stocks: NIFTY 50, Reliance, TCS, Infosys, etc. •Time Range: (e.g., 2015–2023) •Performance Metrics: •MAPE, RMSE, R² formulas •Accuracy tables or graphs (optional screenshots) •Libraries Used: •TensorFlow/Keras, NeuralProphet, NumPy, Pandas, Matplotlib •Model Architecture: •Layers in LSTM (e.g., input → LSTM → Dense) •NeuralProphet configuration (lags, regressors) •Additional Notes: •Limitations and scope for future improvement •Deployment was handled in Streamlit
- 13. Thank you for your time and attention 14th International Conference on Recent Challenges In Engineering And Technology (ICRCET-2025)