Week of October 1, 2024

• Started initial research on financial forecasting and reviewed related literature.
• Began exploring SEC EDGAR API for fetching financial datasets.
• Participated in project meeting with Atakan Erdem.
• Attended seminar: "Introduction to CS491-2 Senior Design Project" by Selim Aksoy.

Week of October 7, 2024

• Encountered problematic issues with SEC EDGAR API due to inconsistent data and documentation.
• Started searching for alternative sources and datasets.
• Participated in weekly meeting with Prof. Altay Güvenir.
• Attended seminar: "Atakan Erdem and Mert Bıçakçı Introduction" by Atakan Erdem and Mert Bıçakçı.

Week of October 14, 2024

• Discovered a Kaggle dataset covering NASDAQ financial data from 2008–2023.
• Performed initial analysis to assess data completeness and quality.
• Participated in weekly meeting with Prof. Altay Güvenir.
• Attended seminar: "Innovation Life Cycle" by Atakan Erdem.

Week of October 21, 2024

• Identified major data gaps and inconsistencies in Kaggle dataset, traced back to original SEC EDGAR API issues.
• Began extensive data cleaning and preprocessing.
• Attended seminar: "The Role of Documentation in (OO) Software Development" by Uğur Doğrusöz.

Week of November 1, 2024

• Continued intensive data cleaning and validation to create a robust dataset for net income loss predictions.
• Participated in weekly meeting with Prof. Altay Güvenir.
• Attended seminar: "La Chanson de Shannon: Shannon's Song" by Fazlı Can.

Week of November 4, 2024

• Successfully established a somewhat reliable dataset structure for net income prediction.
• Started investigating Kaggle competitions related to financial forecasting.

Week of November 11, 2024

• Joined a Kaggle competition focused on financial predictions to gain insights into effective modeling techniques.
• Experimented with initial feature engineering approaches.
• Participated in weekly meeting with Prof. Altay Güvenir.
• Attended seminar: "Tools and Processes in Software Development Lifecycle" by Murat Ergun.

Week of November 18, 2024

• Explored various tabular data models including Random Forest, XGBoost, LightGBM, and CatBoost.
• Evaluated their preliminary performance on financial forecasting tasks.
• Attended seminar: "Technology Entrepreneurship and Investment Ecosystem" by Numan Numan.

Week of December 1, 2024

• Started studying multimodal neural network architectures for potential integration of textual data from reports with tabular data.
• Investigated relevant literature and existing implementations.
• Participated in weekly meeting with Prof. Altay Güvenir.
• Attended seminar: "AI-Driven Mobile Apps: Building Profitable Solutions on a Budget" by Melih Gurgah.

Week of December 8, 2024

• Explored text-tabular and image-tabular multimodal architectures, focusing particularly on text-tabular designs.
• Began drafting initial architecture designs suitable for our data.
• Attended seminar: "Computational Law" by Dilek Küçük.

Week of December 15, 2024

• Implemented a basic multimodal text-tabular neural network.
• Identified issues with handling multi-chunk report data.
• Participated in weekly meeting with Prof. Altay Güvenir.
• Attended seminar: "Bitcoin's First ZK Rollup" by Murat Karademir and Ömer Talip Akalın.

Week of December 22, 2024

• Revised dataset structure and prepared documentation for smoother onboarding and future work.
• Participated in project retrospectives and planning sessions for upcoming semester.
• Participated in weekly meeting with Prof. Altay Güvenir and project meeting with Atakan Erdem; attended senior design seminar.

Week of December 30, 2024

• Finalized semester work; conducted dataset quality assurance checks.
• Prepared initial presentations and reports summarizing semester progress.
• Participated in weekly meeting with Prof. Altay Güvenir and project meeting with Atakan Erdem; attended senior design seminar.

Week of January 1, 2025

• It was final period + spring break, we take a break on project for this period.

Week of February 1, 2025

• Realized dataset was insufficiently rich and complex for optimal modeling performance; began considering strategies for dataset enhancement.
• Continued standard machine learning work to better understand model limitations.
• Participated in weekly meeting with Prof. Altay Güvenir.

Week of February 8, 2025

• Began collaborating with Alara to improve data acquisition using SEC EDGAR API, significantly enriching the dataset.
• Downloaded financial reports from 2001 to 2025 to improve dataset quality.
• Participated in weekly meeting with Prof. Altay Güvenir.

Week of February 15, 2025

• Started part-time role as an LLM researcher, limiting project progress temporarily due to simultaneous work commitments.

Week of February 22, 2025

• Continued balancing new role responsibilities with minimal project work.

Week of March 1, 2025

• Began integrating LLM and chunking strategies into the data pipeline to extract structured financial data directly from textual reports.
• Participated in weekly meeting with Prof. Altay Güvenir.

Week of March 8, 2025

• Successfully enriched dataset with chunked text data, extending dataset coverage back to 2001.
• Encountered challenges with missing data points, began manual correction and data filling process.

Week of March 15, 2025

• Completed painful manual dataset enrichment; initiated training with updated dataset.
• Noted performance improvement in predictive models over baseline heuristics (latest quarter, latest year).
• Participated in weekly meeting with Prof. Altay Güvenir.

Week of March 22, 2025

• Implemented extensive hyperparameter optimization for predictive models, faced difficulties due to financial data characteristics.
• Experimented extensively with various validation schemas to find an effective evaluation strategy.

Week of April 1, 2025

• Selected validation strategy of quarterly training with validation on 2023 data, showing good predictive correlation with 2024.
• Participated in weekly meeting with Prof. Altay Güvenir.

Week of April 8, 2025

• Evaluated models: Random Forest, XGBoost, LightGBM, and CatBoost. Found Random Forest consistently outperformed others.
• Identified overfitting issues with boosting methods despite high regularization.

Week of April 15, 2025

• Finalized Random Forest model achieving a 15% better mean absolute error (net income) and 30% better RMSE (earnings per share) compared to baselines.
• Participated in weekly meeting with Prof. Altay Güvenir.

Week of April 22, 2025

• Prepared final model documentation and started compiling results for project presentation.

Week of April 28, 2025

• Reviewed final model and performance metrics, ensuring readiness for final presentation and report submission.
• Participated in final project review meeting with Prof. Altay Güvenir.