Linear regression is designed for predicting continuous values, not categorical outcomes. Here's a breakdown of the issues:

⚠️ 1. Inappropriate Output Range

• Linear regression outputs any real number, from −∞\infty to +∞+\infty.
• For binary classification, we need outputs between 0 and 1, to represent probabilities.

To compensate, a threshold (e.g., 0.5) is chosen:

• Output ≥ 0.5 → class 1
• Output < 0.5 → class 0

But this introduces problems:

📉 2. Poor Decision Boundary with More Data

• With limited data, a best-fit line might seem to work.
• As more data is added (e.g., outliers or new samples further along the x-axis), the line shifts.
  • This shift moves the threshold.
  • Correct classifications may now become incorrect.

➡️ Model performance degrades with data expansion.

🔁 3. Linear Assumptions Don't Match Classification Needs

• Linear regression tries to minimize squared error.
• But classification aims to minimize classification error (i.e., misclassifications).
• These are not the same objective, so linear regression is not optimized for classification accuracy.