Projects

My expertise in data showcases my ability to extract valuable insights from complex datasets and drive data-informed decision-making. Throughout my career, I’ve worked on a diverse range of projects utilizing cutting-edge technologies and methodologies. From developing predictive machine learning models in Python and R to crafting intricate SQL queries for data manipulation, my work demonstrates a comprehensive approach to data analysis. I’ve also created dynamic visualizations and interactive dashboards using industry-standard BI tools such as Power BI and Tableau, translating raw data into actionable business intelligence. Below, you’ll find a selection of my most impactful projects, highlighting my proficiency in statistical analysis, predictive modeling, and data visualization across various industries and applications.

Multilingual Customer Support Ticket Classification Using SVM

This project classifies multilingual customer support tickets using Support Vector Machines (SVM) to predict ticket types.

Key Steps:

• Data Collection: Downloaded a multilingual customer support ticket dataset from Kaggle.
• Data Cleaning & Transformation: Preprocessed text data using TF-IDF, one-hot encoded categorical features, and balanced classes with SMOTE.
• Model Training: Trained SVM models for English and German ticket classification.
• Model Evaluation: Assessed model performance using accuracy, precision, recall, and F1-score.
• Analysis: Compared performance across languages and identified areas for improvement.

Technologies: Python, scikit-learn, pandas, imblearn, matplotlib, seaborn

Key Insight: SVMs can effectively classify customer support tickets, with performance varying by language due to data characteristics and linguistic complexities. Further improvements are needed to enhance cross-linguistic generalizability.

E-commerce Sales Analysis

This project classifies emails using a Multinomial Naive Bayes model to predict labels (e.g., spam or not spam).

Key Steps:

• Data Collection: Imported sales data and prepared it for analysis in Power BI.
• Data Cleaning & Transformation: Processed missing values, standardized formats, and created calculated fields.
• Visualization & Analysis: Built interactive dashboards to analyze sales trends, customer behavior, and product performance.
• Insights Extraction: Identified seasonal sales patterns, market expansion opportunities, and pricing strategies.
• Recommendations: Proposed data-driven strategies for inventory optimization, targeted marketing, and revenue growth.

Technologies: Power BI, SQL, Excel

Key Insight: Utilizes a probabilistic model to classify emails, with room for improvement in handling class imbalance and model accuracy.

Banknote Authentication Using PCA and KNN

This project uses K-Nearest Neighbors (KNN) with PCA for banknote authentication, optimizing accuracy and fraud detection.

Key Steps:

• Dimensionality Reduction: Applied PCA to extract key features related to banknote dimensions.
• Model Selection: Identified the best k=6 for KNN based on recall optimization.
• Training & Testing: Classified banknotes into genuine or counterfeit using KNN.
• Evaluation: Achieved 99% accuracy, with perfect recall for counterfeit detection.

Technologies: Python, Scikit-learn, PCA, KNN, Pandas, Numpy, Google Colab

Key Insight: The model effectively detects counterfeit banknotes with near-perfect recall, minimizing false negatives in fraud detection.

Email Classification Using Multinomial Naive Bayes

Key Steps:

• Data Collection: Loads email dataset and prepares text data.
• Preprocessing: Cleans text by removing stopwords and applying stemming.
• Feature Extraction: Converts text into numerical features using CountVectorizer.
• Model Training: Trains the Multinomial Naive Bayes model on the data.
• Evaluation: Assesses model performance using a classification report.

Technologies: Python, Pandas, NLTK, Scikit-learn, Google Colab

Key Insight: Utilizes a probabilistic model to classify emails, with room for improvement in handling class imbalance and model accuracy.

Value at Risk (VaR) Analysis of AAPL Stock

This project analyzes the Value at Risk (VaR) of Apple Inc. (AAPL) stock using historical, parametric, and Monte Carlo methods.

Key Steps:

• Data Collection: Fetches AAPL stock prices from Yahoo Finance and calculates log returns.
• VaR Calculation: Computes historical and parametric VaR at a 95% confidence level.
• Monte Carlo Simulation: Generates 10,000 return simulations to estimate risk.
• Backtesting: Counts exceptions to evaluate VaR model accuracy.
• Visualization: Plots return distribution with VaR thresholds.

Technologies: R, quantmod, PerformanceAnalytics, ggplot2, dplyr,

Key Insight: A comparative analysis of different risk estimation techniques to assess potential losses.

Descriptive Statistics and Data Visualization of MPG Dataset

This project explores the MPG dataset through descriptive statistics and visualization techniques to analyze acceleration and other key attributes.

Key Steps:

• Data Collection: Loads and cleans the MPG dataset to ensure accurate analysis.
• Descriptive Statistics: Computes mean, standard deviation, skewness, kurtosis, and other statistical measures.
• Histogram Analysis: Plots acceleration distribution with KDE overlay and statistical annotations.
• Boxplot & Grouped Histograms: Visualizes acceleration across different cylinder configurations.

Technologies: Python, pandas, matplotlib, seaborn, scipy

Key Insight: Statistical and visual exploration of vehicle acceleration patterns, highlighting distribution characteristics and differences by engine type.

Back