Azure AI Engineer Training - Microsoft Azure AI Engineer Training
                     Bias and Variance in 
Machine Learning
Bias and Variance in 
Machine Learning
• Title: Bias and Variance in Machine Learning
• Subtitle: Understanding Model Performance and 
Error
Include your name, date, or other relevant 
information.
Introduction
• Definition of Bias: Bias refers to the error due to overly 
simplistic assumptions in the learning algorithm.
• Definition of Variance: Variance refers to the error due to 
the model's sensitivity to small fluctuations in the training 
data.
• Goal of Machine Learning: Minimize both bias and variance 
to achieve optimal performance.
Bias-Variance Trade-
off
• Explanation: Balancing bias and variance is key in building a 
good model.
– High Bias: Leads to under fitting.
– High Variance: Leads to overfitting.
• Trade-off Illustration: Show a graph that visually explains 
the trade-off.
High Bias (Under fitting)
• Characteristics:
– Simple models (e.g., linear regression)
– Misses important patterns in the data.
– Results in high training and test errors.
• Example: Visual representation of under fitting on a dataset 
(linear model on non-linear data).
High Variance 
(Overfitting)
• Characteristics:
– Complex models (e.g., deep neural networks).
– Captures noise along with the signal.
– Low training error but high test error.
• Example: Visual representation of overfitting (model tightly 
hugging training data points).
Optimal Model (Balanced Bias and 
Variance)
• Characteristics:
– Strikes a balance between bias and 
variance.
– Low training and test error.
– Generalizes well to new data.
• Example: Visual showing a model that fits the 
data appropriately.
Bias-Variance Decomposition
• Formula:
Total Error = Bias² + Variance + Irreducible Error
• Explanation: Breaking down the components 
of model error.
• Graphical Representation: Show how the error 
behaves with increasing model complexity.
Strategies to Handle Bias and Variance
• Reduce Bias:
– Use more complex models.
– Increase model capacity (e.g., from linear regression to 
polynomial regression).
• Reduce Variance:
– Use techniques like cross-validation, regularization (L1/L2), 
and simplifying models.
– Increase training data.
• Practical Example: Briefly describe how these strategies work 
in real-world scenarios.
Conclusion
• Key Takeaways:
– Balancing bias and variance is critical for a well-performing 
model.
– Understand the trade-off to avoid under fitting or 
overfitting.
– Use appropriate techniques to optimize models.
• Closing Thought: In machine learning, the best models aren't 
always the most complex—they are the ones that generalize 
well to unseen data.
CONTACT
Azure AI - 102
Address:- Flat no: 205, 2nd Floor,
Nilagiri Block, Aditya Enclave,
Ameer pet, Hyderabad-1 
Ph. No: +91-9989971070 
Visit: www.visualpath.in 
E-Mail: [email protected]
THANK YOU
Visit: www.visualpath.in