ICT583 Data Science Applications

Directly from the official review slides:

• Binomial distribution is defined by: n (trials) and p (probability of success)
• 68% of data falls within: 1 standard deviation of the mean
• Parameter vs Statistic: parameter = population, statistic = sample
• Null hypothesis: no difference or no relationship between variables
• Residual = vertical distance from data point to regression line
• Gradient descent helps in: minimizing the cost function
• High learning rate: algorithm may overshoot and fail to converge
• Big lambda in regularization: emphasizes small parameters, avoids overfitting
• Sigmoid in logistic regression: models probability that outcome belongs to a class
• Loss function in logistic regression: Cross-Entropy Loss (Log Loss)
• Occam's Razor: simplest model that fits the data is preferable
• Bias-Variance Tradeoff: balance between underfitting and overfitting
• AdaBoost weight increase: classified incorrectly by weak classifier in round t (Answer A)
• Support vectors: data points closest to the decision boundary
• 3D charts disadvantage: distort data and mislead viewers
• Primary goal of model building: generalizes well to unseen data
• K-means NOT required step: None of the above (all steps are required)
• Unsupervised learning: find groups based on similarities with no labels