Event Driven Based Forecasting

Event Driven Based Forecasting is a financial AI project developed at Ozyegin University, focused on ETF price direction prediction and market news sentiment analysis.

Repository: github.com/muzaffercanan/EventDrivenBasedForecasting

Financial prediction requires combining structured market indicators with unstructured news signals while evaluating results through realistic backtesting.

I developed deep learning models, sentiment analysis ensembles, technical indicator visualization workflows, and financial evaluation tooling.

• Python
• TensorFlow/Keras
• PyTorch
• Hugging Face Transformers
• WandB
• FinBERT, DistilBERT, RoBERTa

• Explored ViT, ConvMixer, and MLP-Mixer architectures for technical indicator visualizations.
• Combined multiple transformer models with weighted averaging for SP500 news sentiment.
• Built a backtesting framework with financial evaluation metrics.

• Avoiding overfitting in a noisy financial setting.
• Aligning news sentiment with market movement windows.
• Comparing model results with financial metrics instead of only ML metrics.

The project demonstrates a research-oriented AI workflow that connects market data, visual deep learning, transformer sentiment, and backtesting.

Financial AI systems need both model experimentation and domain-aware evaluation. Accuracy alone is not enough if the strategy fails under realistic market assumptions.