Ahmet Yasir Beydili

Customer Analytics using MySQL, Python & Machine Learning

Tools and Technologies:

• SQL/MySQL: schema design, complex queries, clustering, views, joins, data cleaning
• Python: pandas, matplotlib, seaborn, scikit-learn, and possibly Jupyter Notebook
• ML Models: Logistic Regression, Decision Trees, Random Forests
• Tasks: Customer churn prediction, segmentation, behavior forecasting, exploratory data analysis (EDA), KPI dashboards

Key Qualifications & Responsibilities

• Developed a customer analytics pipeline integrating MySQL and Python for real-world retail datasets.
• Cleaned and structured raw Excel data into a relational database, performing advanced SQL operations including multi-table joins, views, groupings, and time-based aggregations.
• Conducted exploratory data analysis (EDA) and developed interactive visualizations using Pandas, and Matplotlib to uncover behavioral patterns and key performance indicators.
• Applied machine learning algorithms (e.g., Logistic Regression, Decision Trees, Random Forests) via scikit-learn to perform customer churn prediction and segment customer clusters.
• Built modular and scalable Python scripts to handle data preprocessing, model training, evaluation, and result interpretation.

Key Qualifications & Responsibilities

• Designed and implemented machine learning models to predict customer churn in the video streaming industry, using anonymized customer behavior data provided by D-Smart.
• Prepared and engineered the dataset by analyzing user metrics such as AverageContentCompleteRatio, LoginsPerDay, UsageDuration, ContentVariety, and UniqueDevices, applying domain-specific thresholds to label churn cases.
• Conducted feature correlation analysis and impact distribution studies to improve model interpretability and reduce redundancy.
• Trained and evaluated multiple models (e.g., Logistic Regression, Decision Trees, Random Forests and Deep Learning Models), focusing on improving performance metrics through iterative tuning.
• Addressed model underfitting and overfitting by adjusting hyperparameters, applying regularization techniques, and using data balancing methods.
• Presented model accuracy comparisons and insights through visual plots, helping guide the overall pipeline for downstream components like recommendation systems and chatbots.

Key Qualifications & Responsibilities

• Developed a Long Short-Term Memory (LSTM) neural network using PyTorch to forecast Microsoft stock prices based on historical market data.
• Preprocessed and normalized time series data, engineered features, and split the dataset into training and testing sets using a sliding window approach.
• Tuned model hyperparameters including the number of hidden layers, learning rate, and batch size to minimize loss and improve predictive performance.
• Mitigated overfitting by implementing dropout layers and early stopping, and monitored training dynamics through visualization of loss curves.
• Performed backpropagation through time (BPTT) for model optimization and implemented saving/loading functionality for reproducibility and deployment readiness.
• Visualized actual vs. predicted stock prices using Matplotlib, interpreting model results to identify short-term and long-term trends.
• Documented experimental results and technical workflows, ensuring reproducibility and scalability for future use cases.