This project focuses on predicting personal loan acceptance using supervised classification algorithms, specifically K-Nearest Neighbors (KNN) and Logistic Regression. Based on a bank marketing dataset containing customer demographic and financial attributes—such as age, income, education, and account balances—the analysis includes thorough preprocessing, normalization, and feature selection to prepare the data for modeling. Both models were trained and evaluated using accuracy, confusion matrices, and classification reports, with Logistic Regression providing interpretable coefficients and KNN capturing non-linear decision boundaries. The project demonstrates how predictive modeling can support financial institutions in identifying likely loan applicants, enabling more efficient marketing and risk management strategies.

This project tackles the challenge of credit card fraud detection by applying supervised machine learning techniques, specifically Random Forest and K-Nearest Neighbors (KNN). Using a highly imbalanced dataset of anonymized transaction records, the analysis incorporates essential preprocessing steps, including scaling, under-sampling, and handling class imbalance to improve model performance. Both models were evaluated using metrics such as precision, recall, F1-score, and confusion matrices, with a strong focus on minimizing false negatives, critical in fraud detection. Random Forest demonstrated robust performance and interpretability through feature importance analysis, while KNN served as a comparative baseline. This project highlights the application of ensemble and distance-based classifiers in high-stakes, real-world classification tasks involving rare event detection.

This project applies regression techniques to predict the resale price of Toyota Corolla vehicles based on various car features such as age, mileage, fuel type, engine size, and transmission. Utilizing linear regression and exploratory data analysis, the study identifies key predictors influencing car prices and evaluates model performance using metrics like R-squared and RMSE. Feature engineering, outlier removal, and correlation analysis were conducted to improve model accuracy and interpretability. The final model provides actionable insights for dealerships and individual sellers, supporting data-driven decision-making in vehicle pricing and valuation.

This project applies unsupervised machine learning to segment customers based on purchasing behavior using K-Means clustering. Drawing from a retail dataset containing features such as age, annual income, and spending score, the analysis employs exploratory data analysis and visualization techniques to uncover patterns in customer demographics and shopping tendencies. The optimal number of clusters was determined using the Elbow Method, followed by model training and interpretation of the resulting segments. Each cluster represents a distinct customer profile, aiding businesses in tailoring marketing strategies, enhancing customer engagement, and improving retention. This project demonstrates practical use of clustering algorithms for data-driven decision-making in customer analytics.

This project developed a supervised machine learning pipeline to detect toxic language in political discussions on Reddit, integrating TF-IDF and BERT-based embeddings to capture both lexical patterns and contextual nuance. Using a binary classification approach trained on the Jigsaw Toxic Comment dataset and Reddit-specific features, models such as Logistic Regression, XGBoost, and Random Forest were evaluated. The Random Forest classifier achieved the most balanced performance, with 94% accuracy and an F1-score of 0.77. To address class imbalance, SMOTE oversampling was applied. The final model was deployed across over 219,000 Reddit comments, revealing significant regional and ideological disparities in toxicity rates. Insights from this analysis informed the identification of toxic subreddits, influential entities in polarizing discourse, and provided foundational tools for improving online moderation strategies.

This project investigates the impact of socioeconomic and demographic factors—including gender, age, immigrant status, visible minority status, education level, household income, and LGBTQ2+ identity, on mental health outcomes in Canada, using data from the 2022 Mental Health and Access to Care Survey. Through statistical analysis with Python, including chi-squared tests and data visualizations such as heatmaps and mosaic plots, the study identifies significant disparities in the prevalence of mood disorders, anxiety, PTSD, and ADHD across different population groups. Notably, higher rates of mental health disorders were observed among women, LGBTQ2+ individuals, and younger adults. The findings highlight the importance of targeted mental health interventions and provide data-driven insights to support more inclusive and equitable public health policies.

The Shiny app developed for this project offers an interactive platform for exploring a penguin dataset through various visualizations and analyses. Users can delve into exploratory data analysis (EDA) using interactive features like bar charts, scatter plots, and line plots. The app allows for customization, enabling users to select different variables or categories for comparison and interact with the visualizations by hovering over data points or zooming in on specific areas. Through this interactive exploration, users can generate insights and hypotheses about relationships between variables, distribution across categories, and trends within the data.

Conducted a project using Python to implement Latent Dirichlet Allocation (LDA) for topic modeling. The project involved data preprocessing, LDA model training with 20 topics, visualization of topic-word distributions with PyLDAvis, and analysis of topic trends over time. I also addressed challenges faced, compared LDA with other text clustering techniques, and determined the optimal number of topics (15) using Silhouette Scores and Elbow Curves. This project highlighted my expertise in text analysis and demonstrated the practical application of LDA for uncovering topics within a document corpus.

This project focused on analyzing the unemployment rate data for the Portland area over a 20-year period from January 2003 to January 2023. After preprocessing and converting the date column to datetime format, the highest unemployment rate was identified, which occurred in April 2020 at 13.3%. Additionally, a comparison was made between the Portland and national (USA) unemployment rates, revealing similar trends and identifying April 2020 as the period with the highest unemployment rates for both.

This project utilized California house price dataset to predict the price of unseen data based on house features. The project involved data cleaning, visualization, and identifying skewness, followed by the application of different models including Linear Regression, Quadratic Regression, Decision Trees, Random Forests, and Support Vector Machines. The Random Forest model stood out with a lower RMSE, higher R-squared (around 0.866), and lower MAE compared to other models. Overall, the Random Forest model emerged as the most effective in predicting house prices among the evaluated models in this project and used for the final prediction.

Used the Cat and Dog dataset to develop a neural network based on the AlexNet architecture. Employing tools like Keras, TensorFlow, and scikit-learn, the project achieved a remarkable 92% accuracy on unseen data. The AlexNet model, with its 8-layer deep convolutional neural network, excelled in feature extraction. Challenges were addressed through techniques like data augmentation and hyperparameter tuning. Visual aids, including accuracy and loss plots, provided insights into the model's learning process. The final evaluation showcased the model's effectiveness, emphasizing its robust performance in image classification.

Utilized the "Major Crime Indicators" (MCI) dataset from the Toronto Police Service to analyze crime patterns in Toronto from 2014 to 2023. By employing Python for data preprocessing and Tableau for visualization, the project aims to offer actionable insights for law enforcement, policymakers, and local communities to enhance public safety strategies. Through interactive maps and charts generated by Tableau, users can explore crime distribution across neighborhoods, with each crime category represented by a distinct color. The analysis provides a deeper understanding of crime patterns, facilitating targeted strategies against major crimes in the city.

Utilized the "Employee Database.xls" file in Microsoft Power BI, the project visualized the distribution of occupations across genders within the company. The analysis revealed a stark gender imbalance, with men holding a higher number of positions in nearly every occupation, except for Clerical Support roles. Men accounted for approximately 70% of all occupations, while women represented only around 30%. Particularly notable gender disparities were observed in roles such as product assemblers, product testers, Electronics Technicians, and Executives, where men significantly outnumbered women.

Used Qlik's platform to analyze films available on Netflix dataset, showcasing the proportion of genres. Drama and Action/Adventure emerged as the most prevalent genres, demonstrating their popularity within the Netflix audience. This project underscores Qlik's effectiveness in extracting meaningful insights from data to drive informed decision-making and enhance overall performance.

Associated with My internship in Department of National Defence: Designed and implemented Power Apps-based solutions for HR and IT departments, accelerating new employee data entry by 80% and reducing IT service request processing time by 80%. Improved data integrity by migrating to SharePoint lists and enhanced usability with Power BI visualizations, leading to better insights into HR metrics and IT efficiency.

Associated with My internship in Department of National Defence: Created Power Apps applications to streamline boot reimbursement requests and manage medical appointments, reducing wait times by up to 90% and simplifying approval workflows by 80%. Integrated data with SharePoint and automated email updates, enhancing overall efficiency and data integrity.

Associated with My internship in Department of National Defence: Leveraged Power Automate to streamline tasks such as automated emails, SharePoint updates, and ATI request processes, achieving a 95% reduction in manual effort and an 80% improvement in processing time, significantly boosting productivity and user satisfaction. Developed an application for position change requests, automating data extraction from PDFs into SharePoint lists. This reduced human errors by 90% and expedited approval processes, leading to faster decision-making and streamlined workflows.

Associated with My internship in Department of National Defence: Developed numerous Power BI reports and dashboards, utilizing DAX, Power BI Desktop, and Power BI Service. Extracted data from semantic models linked to SharePoint lists, providing clear, actionable insights that improved decision-making across departments.