Assignment 2 — Predictive Modeling Project

---

Project Overview

This project follows the full machine-learning lifecycle required by the assignment:

1. Dataset Selection & Task Definition
   We will choose one dataset (public or provided), describe its context, report statistics, and define a clear predictive task appropriate for the course.
   (Examples: classification, regression, ranking, recommendation, etc.)

2. Data Cleaning & Exploratory Analysis (EDA)

   • Inspect raw data
   • Handle missing values, duplicates, and type conversions
   • Generate descriptive statistics
   • Visualize distributions, correlations, and feature patterns

3. Feature Engineering & Modeling

   • Prepare inputs for ML models (numeric, categorical, text, etc.)
   • Implement course-relevant baseline models (e.g., Logistic Regression, Naive Bayes, Linear Regression)
   • Build improved/advanced models (SVM, Random Forest, XGBoost, Matrix Factorization, etc.)
   • Log all model runs to MLflow/DagsHub for experiment tracking

4. Evaluation & Comparison

   • Choose evaluation metrics appropriate for the predictive task
   • Compare baselines vs. advanced models
   • Visualize results (confusion matrices, ROC curves, feature importances, etc.)
   • Select best model based on MLflow-tracked metrics

5. Discussion & Related Work

   • Review how similar datasets or tasks have been studied in prior work
   • Compare our results with published findings when applicable
   • Discuss limitations, failure cases, and potential improvements

---

MLflow Experiment Tracking (via DagsHub)

To maintain reproducibility and manage model experimentation cleanly, we use DagsHub’s built-in MLflow tracking.

Logged information includes:

• Model hyperparameters
• Training/validation metrics
• Feature configurations
• Model artifacts
• Best-performing run for each model type

This allows the group to collaborate across machines (local, Colab, Kaggle, or Linux server) while keeping all experiment artifacts synchronized.