Yusuf Erdem Altinsoy

In [1]:

from portfolio import erdem

In [2]:

erdem.describe()

Out[2]:

Computer Engineering student at Yeditepe University focused on machine learning and data analysis. Builds gradient boosting models for fuel-efficient race strategy at Seven Racing, reached the top 6% in Kaggle's Spaceship Titanic, and designs data pipelines for football analytics. Seeking a data science internship.

	erdem
name	Yusuf Erdem Altinsoy
university	Yeditepe University
degree	B.Sc. Computer Engineering
period	Sept. 2023 – expected June 2028
location	Istanbul, Turkey
seeking	data science internship

In [3]:

erdem.projects

Out[3]:

	title	status	stack	summary
race-strategy	Race strategy model	In progress	XGBoost, LightGBM, Pandas	Gradient boosting models for Seven Racing that decide, sector by sector, whether the car should accelerate or coast to maximize fuel efficiency.
fbref-scraper	fbref_scraper	Built	Python 3.11, nodriver, BeautifulSoup, pydantic, SQLAlchemy, PostgreSQL	A modular fbref.com football data scraper: 4 page types parsed by 19 scraper modules into 26 typed models with 118 validated fields.
kaggle-competitions	Kaggle competitions	Built	Pandas, scikit-learn, XGBoost, LightGBM, CatBoost	Spaceship Titanic (solo): rank 95 of 1,583 (top 6%), 0.810 public accuracy against a roughly 50% majority-class baseline.

In [4]:

dir(erdem)

Out[4]:

['describe()', 'info()', 'projects', 'experience', 'skills', 'contact']

Type erdem.info() or any other name from the list into the cell below and press Enter, or click a name.

In [ ]:

In [5]:

erdem.describe()

Out[5]:

	erdem
name	Yusuf Erdem Altinsoy
university	Yeditepe University
degree	B.Sc. Computer Engineering
period	Sept. 2023 – expected June 2028
location	Istanbul, Turkey
seeking	data science internship

In [5]:

erdem.info()

<class 'portfolio.Erdem'> Education B.Sc. Computer Engineering, Yeditepe UniversityIstanbul, TurkeySept. 2023 – expected June 2028English-medium, including a prep year Skills Languages: Python, SQL (PostgreSQL), C++, CTechnologies: Pandas, NumPy, scikit-learn, XGBoost, LightGBM, CatBoost, BeautifulSoup, nodriver, Jupyter, Git Spoken languages Turkish: NativeEnglish: B2. University preparatory program; degree taught fully in English; three months living abroad in Germany.

In [5]:

erdem.projects

Out[5]:

	title	status	stack	summary
race-strategy	Race strategy model	In progress	XGBoost, LightGBM, Pandas	Gradient boosting models for Seven Racing that decide, sector by sector, whether the car should accelerate or coast to maximize fuel efficiency.
fbref-scraper	fbref_scraper	Built	Python 3.11, nodriver, BeautifulSoup, pydantic, SQLAlchemy, PostgreSQL	A modular fbref.com football data scraper: 4 page types parsed by 19 scraper modules into 26 typed models with 118 validated fields.
kaggle-competitions	Kaggle competitions	Built	Pandas, scikit-learn, XGBoost, LightGBM, CatBoost	Spaceship Titanic (solo): rank 95 of 1,583 (top 6%), 0.810 public accuracy against a roughly 50% majority-class baseline.

In [5]:

erdem.experience

Out[5]:

Software Team Member, Yeditepe University Seven Racing

Istanbul, Turkey, Sept. 2025 – present

Strategy modeling. Developing gradient boosting models (XGBoost, LightGBM) that decide, sector by sector, whether the car should accelerate or coast to maximize fuel efficiency, trained on optimal decisions derived from physics-based calculations.
Telemetry features. Engineered model inputs with Pandas from live vehicle telemetry (fuel level, engine state, temperature, track position) captured by an on-board Raspberry Pi 5 and logged to CSV.
Predictive planning. Incorporated upcoming track features such as corners and uphill sections so the strategy plans 1–2 sectors ahead instead of reacting only to the current state.

In [5]:

erdem.skills

Out[5]:

{'Languages': ['Python', 'SQL (PostgreSQL)', 'C++', 'C'], 'Technologies': ['Pandas', 'NumPy', 'scikit-learn', 'XGBoost', 'LightGBM', 'CatBoost', 'BeautifulSoup', 'nodriver', 'Jupyter', 'Git']}

In [5]:

erdem.contact

Out[5]:

{'email': 'altinsoyyusuferdem@gmail.com', 'github': 'https://github.com/erdemalti0', 'linkedin': 'https://linkedin.com/in/yusuf-erdem-altinsoy', 'kaggle': 'https://kaggle.com/yusuferdemaltinsoy'}

In [5]:

erdem.projects.loc["race-strategy"]

Out[5]:

Race strategy model

status	In progress
stack	XGBoost, LightGBM, Pandas

This work is in progress. No results have been published.

Work done as a Software Team Member at Yeditepe University Seven Racing, Sept. 2025 – present.

Strategy modeling

Developing gradient boosting models (XGBoost, LightGBM) that decide, sector by sector, whether the car should accelerate or coast to maximize fuel efficiency, trained on optimal decisions derived from physics-based calculations.

Telemetry features

Engineered model inputs with Pandas from live vehicle telemetry (fuel level, engine state, temperature, track position) captured by an on-board Raspberry Pi 5 and logged to CSV.

Predictive planning

Incorporated upcoming track features such as corners and uphill sections so the strategy plans 1–2 sectors ahead instead of reacting only to the current state.

In [5]:

erdem.projects.loc["fbref-scraper"]

Out[5]:

fbref_scraper

status	Built
stack	Python 3.11, nodriver, BeautifulSoup, pydantic, SQLAlchemy, PostgreSQL
repository	github.com/erdemalti0/fbref_scraper

What it is

A modular scraper for football data on fbref.com.

What it parses

4 page types (match, player, club-season, league) are parsed by 19 scraper modules into 26 typed models with 118 validated fields.

Storage

Reports are stored as JSON and/or upserted into PostgreSQL JSONB tables keyed on page ID, so re-scrapes update rows instead of duplicating them.

In [5]:

erdem.projects.loc["kaggle-competitions"]

Out[5]:

Kaggle competitions

status	Built
stack	Pandas, scikit-learn, XGBoost, LightGBM, CatBoost

	Spaceship Titanic
cv accuracy	0.811
public accuracy	0.810
rank	95 / 1,583

Spaceship Titanic

Solo entry. Rank 95 of 1,583 (top 6%), with 0.810 public accuracy against a roughly 50% majority-class baseline. Public accuracy rose from 0.797 to 0.810 over iterative submissions.

Models

Gradient boosting classifiers and regressors with XGBoost, LightGBM and CatBoost, built on Pandas-based data cleaning and feature engineering.

Tuning and validation

GridSearchCV over 52 LightGBM and CatBoost configurations (learning rate, tree depth, number of leaves; 260 fits), scored by stratified 5-fold cross-validation.

The selected model's 81.1% CV accuracy closely matched its 0.810 public score, indicating the choice was not tuned to the leaderboard.

Other competitions

Also entered Titanic (classification) and House Prices (regression).