Blog

Built a distributed movie recommendation engine using Spark MLlib on AWS EMR to analyze the MovieLens 10M dataset (10 million ratings from 72,000 users across 10,000 movies). Leveraged MLlib…

I explored the Image Segmentation dataset from the UCI machine learning repository, which describes 2,310 segmented regions (e.g., brick, sky, foliage, window) with 19 numerical attributes. The…

Using the modified Jester online joke ratings dataset, I built an item-based recommender that suggests jokes to users based on how similarly other users rate the same jokes. The dataset contains…

Built a distributed movie recommendation engine using Apache Spark on AWS EMR to analyze the MovieLens 10M dataset (10 million ratings from 72,000 users across 10,000 movies). Implemented…

Implemented Google's PageRank algorithm from scratch using Python and pandas, demonstrating how web pages are ranked based on link structure. Built custom transition matrix creation and power…

Explored Apache Cassandra's CQL query patterns and Apache Storm's windowing mechanisms for real-time stream processing. Demonstrated Cassandra's limitations with ORDER BY (only works on…

Applied K-means clustering to 2,225 BBC News articles (2004-2005) using term frequency features. K=5 achieved 76.4% homogeneity and 76.7% completeness against editorial categories (business…

Built comprehensive regression pipeline predicting violent crime rates across 1,994 US communities using 97 socioeconomic features (census + FBI data). Lasso achieved best test RMSE of 0.143…

Deployed HBase NoSQL database using Docker Compose with Master, RegionServer, and Zookeeper components. Demonstrated column-oriented key-value storage operations including table creation, data…

Built a distributed movie recommendation engine using MapReduce (mrjob 0.6.12) on AWS EMR, processing 1 million ratings from the MovieLens dataset. Computed cosine similarity scores between…

Implemented custom KNN and Rocchio classifiers for newsgroup document classification (800 training, 200 test, 5,500 features). Rocchio method achieved 71% accuracy outperforming KNN (67% with…

This comprehensive tutorial covers exploratory data analysis and preprocessing techniques using the Adult Census dataset (32K+ records). Key topics include: handling missing values with…

Built a complete 4-stage ML pipeline to predict a developer's programming language choice using binary classification: (1) Cross-validation for model evaluation, (2) Grid search for automatic…

Implemented Markov Decision Process (MDP) and value iteration to solve the FrozenLake 8x8 environment, achieving 100% success rate compared to 0.88% for random policies (113x improvement…

Built end-to-end ML pipeline combining StandardScaler → PCA → KNeighborsClassifier, achieving 95% accuracy on Iris dataset. Used GridSearchCV to optimize entire pipeline simultaneously…

Implemented and compared three AI search algorithms (BFS, UCS, A * ) for delivery route optimization on real-world map data from Tegucigalpa (24 locations, 64 streets). A * dramatically…

Applied PCA to reduce Iris dataset from 4D to 2D while retaining 95.8% variance (PC1: 72.77%, PC2: 23.03%). Demonstrates dimensionality reduction fundamentals with bivariate visualization of…

Compared t-SNE, UMAP, and PCA for dimensionality reduction and visualization of high-dimensional data (500 samples, 4 clusters). UMAP best preserves global structure with clearer cluster…

Compared DBSCAN vs HDBSCAN for clustering Canadian museum locations (250+ samples). DBSCAN identified 6 clusters (eps=2, min_samples=5) with 3.6% noise, while HDBSCAN discovered…

Compared Decision Tree vs SVM for detecting fraud in 284K+ credit card transactions with extreme class imbalance (99.8% legitimate). Used undersampling to balance classes, achieving Decision…

Compared softmax regression vs One-vs-Rest and One-vs-One strategies for multi-class Iris classification (3 species, 150 samples). Softmax achieved 94.7% test accuracy matching One-vs-Rest…

Developed a logistic regression model to predict telecom customer churn with 80% accuracy and 0.71 F1-score for churn class. Analyzed 200 customer records with features including tenure, age…

Automated hyperparameter optimization for SVM on Iris dataset, testing 2,700 parameter combinations (3 kernels × 9 C values × 10 gamma values × 10 CV folds). Found optimal configuration: RBF…

Implemented K-Means clustering for customer segmentation, identifying 3 optimal clusters using elbow method analysis (850 customers). Segments include education-focused (avg age 42, education…

Built an interpretable Decision Tree classifier for patient drug prescription with 95% accuracy (200 patient records). Model predicts Drug Y with 100% precision, Drug X with 91.7% precision…

Built an SVM classifier to predict benign vs malignant cancer cells with 95.71% accuracy on test data (683 samples). Used RBF kernel for non-linear classification, achieved 100% precision for…

Compared simple and multiple linear regression for predicting vehicle CO2 emissions (1,067 samples). Multiple regression with engine size, cylinders, and fuel consumption achieved R² = 0.8…

Built Random Forest regressor for California housing price prediction (20,640 samples, 8 features) achieving R² = 0.81 on test set (MAE: $33K, RMSE: $49K). Analyzed model performance across…

Deployed Node.js applications to Fly.io's global edge network with automatic HTTPS, demonstrating modern serverless deployment workflow. Fly.io offers free tier with 3 VMs, automatic scaling…

Explored Bitcoin Lightning Network as Layer 2 scaling solution enabling instant, low-fee micropayments off-chain. Lightning achieves millions of transactions per second vs Bitcoin's 7 TPS…