0% found this document useful (0 votes)
15 views4 pages

AI and Machine Learning Concepts Explained

The document provides an overview of artificial intelligence (AI), machine learning (ML), deep learning (DL), and data science (DS), highlighting their definitions, differences, and applications. It discusses various learning methods, including supervised, unsupervised, semi-supervised, and reinforcement learning, along with techniques for handling data issues like overfitting, missing data, and imbalanced datasets. Additionally, it covers feature selection and engineering methods, emphasizing their importance in improving model performance and efficiency.
Copyright
© All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
0% found this document useful (0 votes)
15 views4 pages

AI and Machine Learning Concepts Explained

The document provides an overview of artificial intelligence (AI), machine learning (ML), deep learning (DL), and data science (DS), highlighting their definitions, differences, and applications. It discusses various learning methods, including supervised, unsupervised, semi-supervised, and reinforcement learning, along with techniques for handling data issues like overfitting, missing data, and imbalanced datasets. Additionally, it covers feature selection and engineering methods, emphasizing their importance in improving model performance and efficiency.
Copyright
© All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd

ML Assignment-1

Section-1
1. AI Definition: Simulation of human intelligence in machines to perform tasks like reasoning
and learning.

2. AI vs ML vs DL vs DS:

o AI: Broad field (robotics, NLP)

o ML: Subset where systems learn from data

o DL: Neural networks for complex patterns

o DS: Data analysis + ML

3. AI vs Traditional Software:

o AI adapts; software follows fixed rules

4. Examples:

o AI: Self-driving cars

o ML: Spam filters

o DL: Face recognition

o DS: Customer analytics

5. Importance: Automation, predictive analytics, etc.

6. Supervised Learning: Training on labeled data (input-output pairs).

7. Examples: Linear Regression, SVM.

8. Process: Data → Train → Validate → Test.

9. Unsupervised Learning: Finds patterns in unlabeled data (clustering).

10. Examples: K-Means, PCA.

11. Semi-Supervised: Uses both labeled/unlabeled data.

12. Reinforcement Learning: Learns via rewards (e.g., game AI).

13. RL vs Others: No labels; goal-oriented learning.

Section-2
14. Train-Test-Val Split: Prevents overfitting (70-15-15).

15. Training Set: Teaches model patterns.

16. Split Sizes: Large data: 70-15-15; Small: 60-20-20.

17. Bad Splits Cause: Overfitting/underfitting.


18. Trade-offs: More training = better learning but less testing.

19. Model Performance: Accuracy on unseen data.

20. Metrics: Accuracy, RMSE, F1-Score.

21. Overfitting: Model memorizes noise → poor generalization.

22. Solutions: Dropout, regularization.

23. Underfitting: Too simple → high bias.

24. Fix: Add features, reduce regularization.

25. Bias-Variance Tradeoff: Balance simplicity/complexity.

26. Missing Data Handling: Deletion, imputation (mean/median).

27. Ignoring Missing Data: Biases results.

28. Imputation Pros/Cons:

• Mean: Simple but reduces variance

• KNN: Accurate but slow

29. Impact on Performance: Skews predictions.

30. Imbalanced Data: One class dominates (e.g., 95% non-fraud).

31. Challenges: Model ignores minority class.

32. Solutions: SMOTE, class weights.

33. Up/Down-Sampling:

• Up: Duplicate minority

• Down: Reduce majority

34. SMOTE: Generates synthetic minority samples.

35. Interpolation: Estimates missing values (linear/polynomial).

36. Outliers: Extreme values skewing results.

37. Handling: Remove, transform, or use robust models.

Section-3
46. Feature Selection Methods:

• Filter: Statistical tests (Pearson)

• Wrapper: Uses model performance (RFE)

• Embedded: Built-in (Lasso)

47. Examples:

• Filter: Chi-Square
• Wrapper: Backward Elimination

48. Pros/Cons:

• Filter: Fast but ignores feature interactions

• Wrapper: Accurate but computationally heavy

49. Feature Scaling: Normalizes data (e.g., Min-Max).

50. Standardization: (X - μ)/σ → mean=0, std=1.

51. Mean Normalization: Scales to [-1,1] vs. standardization.

52. Min-Max Scaling: [0,1] range; sensitive to outliers.

53. Unit Vector Scaling: Normalizes to length=1.

54. PCA: Reduces dimensions while preserving variance.

55. PCA Steps:

56. Standardize

57. Covariance matrix

58. Eigenvectors/values

59. Select top components

60. Eigenvalues: Indicate variance explained.

61. Dimensionality Reduction: Projects data to lower dimensions.

62. Data Encoding: Converts categories to numbers.

63. Nominal Encoding: No order (One-Hot).

64. One-Hot Encoding: Binary columns per category.

65. Many Categories: Use hashing or embedding.

66. Mean Encoding: Replaces categories with target mean.

67. Ordinal Encoding: Ordered labels (e.g., small=1, medium=2).

68. Target Guided Encoding: Uses target relationship.

69. Covariance: Measures joint variability.

70. Correlation Check: Pearson/Spearman tests.

71. Pearson: Linear relationships (-1 to 1).

72. Spearman: Monotonic relationships.

73. VIF: Detects multicollinearity (>5 = problematic).

74. Feature Selection Purpose: Improves model efficiency.

75. RFE: Recursively removes least important features.


76. Backward Elimination: Starts with all features, removes one by one.

77. Forward Elimination: Starts empty, adds best features.

78. Feature Engineering: Creates new features (e.g., log transforms).

79. Steps:

80. Domain research

81. Transformations

82. Validation

83. Examples: Binning, polynomial features.

84. Feature Selection vs Engineering:

• Selection: Chooses best existing features

• Engineering: Creates new features

78. Importance: Reduces noise, improves accuracy.

79. Impact on Performance: Better features → better models.

80. Choosing Features: Use domain knowledge + statistical tests.

You might also like