Supervised machine learning
Mobile plan classification
Compare decision trees, random forests, and logistic regression to predict a customer's observed mobile plan from usage data.
Explore the project: Mobile plan classificationBack to the homepage
Educational Projects
Learning through practical projects in data science and machine learning.
A selection from my learning journey into AI engineering. Each project explores a specific problem, method, and lesson.
This selection follows my progress from data exploration to model development and interactive applications.
Exploratory data analysis
Video game sales strategy
Explore platform lifecycles, regional preferences, review scores, and historical sales to support a simulated marketing strategy.
Explore the project: Video game sales strategyStatistical analysis
Mobile plan revenue analysis
Combine customer, call, message, internet, and pricing tables to compare monthly revenue across two prepaid plans.
Explore the project: Mobile plan revenue analysisInteractive data application
Vehicle listings dashboard
A Streamlit application that lets visitors explore odometer readings and asking prices through interactive Plotly charts.
Explore the project: Vehicle listings dashboardSupervised machine learning
Mobile plan classification
Compare decision trees, random forests, and logistic regression to predict a customer's observed mobile plan from usage data.
The question
Use monthly calls, minutes, messages, and data usage to classify customers into Smart or Ultra plans.
Approach
Stratified training, validation, and test splits; tree hyperparameter searches; comparison against a majority-class baseline.
What I learned
The validation-selected random forest outperformed the baseline in the recorded study. The work develops model selection and evaluation skills beyond exploratory analysis.
Study scope
An offline course study. Predicting the current plan does not identify the best plan for a customer. A fresh holdout is needed after further test-set exploration; the notebook also retains a course-environment dataset path.
Exploratory data analysis
Video game sales strategy
Explore platform lifecycles, regional preferences, review scores, and historical sales to support a simulated marketing strategy.
The question
Use sales history through 2016 to identify platforms and audience patterns relevant to a 2017 campaign scenario.
Approach
Data preparation, platform and genre comparisons, regional segmentation, correlations, Kruskal–Wallis testing, and Welch's t-tests.
What I learned
Regional profiles and platform lifecycles change the interpretation of aggregate sales. The analysis connects exploratory findings to a business question.
Study scope
Historical educational scenario, not a current market forecast. The full dataset is not included; the repository provides a sample. Correlation does not establish that reviews cause sales.
Statistical analysis
Mobile plan revenue analysis
Combine customer, call, message, internet, and pricing tables to compare monthly revenue across two prepaid plans.
The question
Investigate how Surf and Ultimate plans differ in customer usage and revenue within a 2018 sample.
Approach
Join five datasets, aggregate activity by customer and month, implement billing rules, visualize distributions, and apply Welch's t-tests.
What I learned
Higher total revenue and higher revenue per customer are different business signals. The study separates those measures and links billing logic to analysis.
Study scope
Educational analysis of a historical sample. Repeated monthly observations from the same customers limit independence assumptions; the results do not demonstrate that an advertising change will increase revenue.
Interactive data application
Vehicle listings dashboard
A Streamlit application that lets visitors explore odometer readings and asking prices through interactive Plotly charts.
The question
Make a vehicle listings dataset easier to explore through a small web interface.
Approach
Load listings with pandas, use a checkbox to display an odometer histogram, and use a button to generate a price-versus-odometer scatter plot.
What I learned
The project turns an analysis into an interactive interface and introduces application deployment, complementing the notebook-based studies.
Study scope
An exploratory dashboard of advertised listings. It does not estimate sale prices, establish causation, or implement a production valuation model.