INITIALIZING DEEP ARCHIVES...
High Utility Itemset Mining (HUIM) is a critical data mining task focused on discovering combinations of items that yield high profit or utility in transactional databases. This research explores MHUIMiner, an efficient algorithm designed to overcome the computational and memory bottlenecks of traditional frequent pattern mining methods.
This project explores the reduction of a spatial Constraint Satisfaction Problem (CSP) into a Boolean Satisfiability (SAT) problem. The objective is to resolve a grid-coloring logic puzzle where local neighborhood constraints dictate the global state of the matrix. By translating the spatial constraints into Conjunctive Normal Form (CNF) and utilizing the CDCL-based (Conflict-Driven Clause Learning) `Glucose3` solver, the system guarantees a mathematically sound resolution or a formal proof of unsatisfiability in optimal time.
This project implements a custom **Naive Bayes Classifier built entirely from scratch** in Python. The objective is to predict student outcomes (Pass/Fail) based on a continuous dataset of sequential quiz scores. Rather than relying on high-level machine learning libraries like `scikit-learn`, this project mathematically constructs the probabilistic model, handling everything from data imputation to numerical stability optimizations.
A scalable time-series forecasting pipeline built with Apache Spark. By leveraging PySpark Window functions for distributed lag-feature engineering and MLlib's Linear Regression, this model accurately predicts gold prices based on a 15-year dataset, achieving an R² of 0.9995.
