Class 10 AI | Unit 2: Advanced Concepts of Modeling in AI — Notes, MCQs, Case Studies & Free Worksheet
AI Logic School · Class 10 · Artificial Intelligence
Unit 2: Advanced Concepts
of Modeling in AI
Comprehensive study notes, MCQs, Q&A, Case Studies & Practice Worksheet — aligned to CBSE board examination pattern
50
MCQs
30
Q&A Pairs
4
Case Studies
30
Worksheet Qs
Study Notes
Unit 2: Advanced Concepts of Modeling in AI — Key concepts for Board Examination
1. What is a Model in AI?
A model in AI is a mathematical representation of a real-world process. It is trained on data to recognize patterns and make predictions or decisions.
Key Formula: Model = Algorithm + Data → Prediction/Output
- Models learn from training data and are tested on test data
- Goal: Generalize well to unseen/new data
- Types: Classification, Regression, Clustering
2. Types of Machine Learning
- Supervised Learning: Model trained with labeled data. Examples: spam detection, image classification.
- Unsupervised Learning: Model finds hidden patterns in unlabeled data. Examples: customer segmentation, anomaly detection.
- Reinforcement Learning: Agent learns by receiving rewards/penalties. Examples: game playing (chess, Go), robotics.
Supervised
Unsupervised
Reinforcement
Semi-Supervised
3. Training, Validation & Testing
- Training Set: Data used to train the model (typically 70–80%)
- Validation Set: Used to tune hyperparameters (typically 10–15%)
- Test Set: Used to evaluate final model performance (typically 10–20%)
⚠ Important: Never use test data during training — it leads to data leakage and inflated accuracy.
4. Overfitting & Underfitting
- Overfitting: Model performs well on training data but poorly on new data. Model memorizes instead of learning.
- Underfitting: Model too simple to capture patterns — poor on both training and test data.
- Good Fit: Balanced performance on both sets.
Solutions to Overfitting: More training data, Regularization (L1/L2), Dropout, Pruning, Cross-validation
5. Bias and Variance
- Bias: Error from oversimplified assumptions. High bias → Underfitting.
- Variance: Error from sensitivity to training data. High variance → Overfitting.
- Bias-Variance Tradeoff: Reducing bias increases variance and vice versa.
High Bias = Underfitting
High Variance = Overfitting
Goal = Low Bias + Low Variance
6. Model Evaluation Metrics
- Accuracy: (Correct Predictions / Total) × 100
- Precision: TP / (TP + FP)
- Recall: TP / (TP + FN)
- F1-Score: Harmonic mean of Precision and Recall
- Confusion Matrix: Table showing TP, TN, FP, FN
- MSE: Mean Squared Error — for regression models
Remember: Accuracy is misleading for imbalanced datasets. Always check Precision, Recall, F1-Score.
7. Neural Networks & Deep Learning
- Inspired by the human brain — composed of neurons arranged in layers
- Input Layer: Receives raw data
- Hidden Layer(s): Learns features/patterns
- Output Layer: Produces prediction
- Activation Functions: ReLU, Sigmoid, Softmax — adds non-linearity
- Backpropagation: Adjusts weights by propagating error backwards
8. Feature Engineering & Data Preprocessing
- Feature Selection: Choosing the most relevant variables
- Normalization: Scaling features to 0–1 range
- Standardization: Mean = 0, SD = 1
- Handling Missing Values: Mean/Median imputation, dropping rows
- Encoding Categorical Data: Label Encoding, One-Hot Encoding
9. Popular AI/ML Algorithms
- Linear Regression: Predict continuous values
- Logistic Regression: Binary classification
- Decision Tree: Tree-based model using feature splits
- Random Forest: Ensemble of decision trees
- KNN: Classifies based on nearest neighbors
- K-Means: Unsupervised grouping into K clusters
- SVM: Finds optimal hyperplane to separate classes
10. AI Project Cycle
- Step 1: Problem Scoping
- Step 2: Data Acquisition
- Step 3: Data Exploration
- Step 4: Modeling
- Step 5: Evaluation
11. Cross-Validation
K-Fold Cross-Validation: Dataset split into K folds. Model trained K times, each using a different fold as validation. Final performance = average of all K runs. Common K = 5 or 10.
12. Transfer Learning
- Using a model trained on one task for a different related task
- Saves time and resources — useful when training data is limited
- Examples: VGG16, ResNet (images), BERT, GPT (language)
13. Ethics in AI Modeling
- Bias: Biased data leads to discriminatory outputs
- Explainability (XAI): Users should understand AI decisions
- Privacy: Models must not expose training data
- Fairness: Models must treat all groups equally
50 Multiple Choice Questions
Click on any option to check your answer. Explanation appears after selection.
Short & Long Answer Questions
Click any question to reveal the model answer. Perfect for board exam preparation.
Section A: 1-Mark Questions
Section B: 2-Mark Questions
Section C: 3-Mark Questions
Section D: 5-Mark Questions
Case-Based Questions
Read each scenario carefully and answer the questions. Aligned to CBSE board pattern.
Practice Worksheet
30 mixed questions with navigation. Track your score as you go!
Practice Worksheet
Question 1 of 30
Score: 0/30
Comments
Post a Comment