Optimizing Model Performance: Baseline Selection and Tuning

Learning Objectives

Diagnose poor model performance (low R²) and identify underlying issues (e.g., multicollinearity, incorrect model assumptions)

Understand the limitations of linear regression in MMM

Address multicollinearity by using appropriate techniques

Improve model accuracy and performance using data transformation and feature engineering

Overview

You're working on a Market Mix Modeling (MMM) pipeline to understand how different advertising spends (TV, Radio, Online) influence sales. The team has provided you with a baseline Linear Regression model. However, the model performance is poor, with low R-squared (R²) and predictions that don’t seem to capture the real-world impact of ad spend on sales.

Your task is to identify the issues in the baseline model, propose a better approach, and improve the model performance.

Prerequisites

Knowledge of Market Mix Modeling (MMM) concepts
Familiarity with Linear Regression and log-transformation
Understanding of multicollinearity and how it impacts regression models
Experience with model evaluation metrics, specifically R²
Knowledge of data preprocessing and correlation analysis