Understanding who has access to green spaces in cities requires more than just maps. In this case study, I used QGIS to transform raw spatial layers: administrative boundaries, building footprints, road networks, and park polygons—into actionable insights about urban accessibility.

The workflow covered geospatial preprocessing, multi-layer map visualization to reveal spatial patterns, and network-based analysis including distance matrices and service-area (iso-area) calculations.

In machine learning, great models start with great data. Data preprocessing is the process of cleaning, transforming, and preparing raw data so machine learning algorithms can learn effectively. This includes fixing incorrect or missing values, removing irrelevant information, and organizing data into a usable format.

By preparing the data properly, we improve accuracy, reduce model bias, and ensure reliable predictions. In short, data preprocessing is a foundational step that cleans messy data and makes machine learning models perform better and more consistently.

This case study demonstrates how structured requirements help guide successful software design. The system adapts learning paths, provides diagnostic and practice tests, gives tailored feedback, and tracks student progress toward a target score. The PRD outlines the product vision, background, key stakeholders, and functional scope while emphasizing the importance of clear requirements in reducing project risks and aligning development teams.

The final prototype includes user flows, feature descriptions, and interface mockups for desktop and mobile. This exercise reinforces how effective requirement documentation creates clarity, improves user outcomes, and strengthens the foundation for scalable solutions.

This project explores how Natural Language Processing (NLP) can be used to automatically detect sentiment in user-generated movie reviews. Using IMDb review data, I built a machine learning model to classify text as positive or negative, addressing the challenge of manually processing large-scale review content.

The workflow included text preprocessing, followed by model training and evaluation using metrics such as F1-score, AUC-ROC, and average precision. The model successfully identified sentiment patterns from real-world text data, demonstrating how NLP can support content platforms in understanding audience reactions at scale, improving decision-making in content curation, recommendation systems, and reputation monitoring.

Matrix multiplication (dot product) is not typically used as a standard data-masking method. However, it can serve as a masking mechanism when applied as a transformation that alters the original values. By multiplying input data with a transformation matrix, the resulting output becomes a mathematically obscured representation of the data. In deep learning workflows, this transformation can act like a lightweight obfuscation step, hiding raw values while still allowing the information to be processed for learning purposes.

This project demonstrates how GridSearchCV can be used to efficiently tune hyperparameters across multiple machine learning models to predict the optimal steel-heating temperature and reduce electricity consumption. Hyperparameter tuning plays a critical role in model performance, similar to adjusting ingredients in a recipe to achieve the best result. By methodically exploring different parameter combinations, GridSearchCV automates the search for the most effective model configuration. The article emphasizes understanding the problem and data before tuning, as well-defined objectives guide meaningful model improvements and ensure machine learning is applied appropriately and effectively.