272 old exam short answer

Supervised vs Unsupervised Learning

Supervised uses labeled data for prediction; unsupervised finds patterns in unlabeled data.

Examples of Supervised Learning

Fraud detection, email spam classification — clear input-output pairs.

Examples of Unsupervised Learning

Customer segmentation, anomaly detection — no predefined labels.

Merging vs Concatenating in pandas

Merging joins DataFrames on keys; concatenating stacks them by axis.

When to use merge in pandas

When combining data from two sources with a common column (e.g., ID).

ANOVA Hypotheses

Null: all group means are equal; Alternative: at least one differs.

When to use ANOVA vs t-test

Use ANOVA for 3+ groups; t-test for 2 groups.

What is an Outlier

A value far from other data points that can distort model accuracy.

Effects of Outliers on Models

Can skew regression lines, increase error, or influence clustering.

SQL vs Text File

SQL databases handle large structured data more efficiently than text files.

SQLite

Lightweight SQL database stored in a single file; great for local analysis.

Categorical Variable

Variable with non-numeric categories (e.g., color, gender, job title).

Encoding Categorical Variables

Use one-hot encoding or label encoding for regression/classification.

Overfitting

Model memorizes training data but fails to generalize; common with small datasets.

Preventing Overfitting

Use cross-validation, regularization, pruning, or dropout.

Principal Component Analysis (PCA)

Reduces dimensionality while preserving variance via uncorrelated components.

Randomization in Ensembles

Introduces diversity in training data or features to reduce overfitting.

Random Forest

Ensemble of decision trees trained on bootstrapped data with random feature subsets.

Classification vs Regression

Classification = categories; Regression = continuous values.

Time Series Components

Trend, seasonality, cyclical, and random components.

Autocorrelation

Correlation of a time series with a lagged version of itself.

Importance of Autocorrelation

Helps identify patterns and predict future values in time series.

R² (coefficient of determination)

Measures how well the regression model explains data variability.

Use of R²

Compares model fit across different regression models.

Random Numbers in Data Analysis

Used in simulation, sampling, and randomized algorithms.

Monte Carlo Simulation

Method using repeated random sampling to model probabilistic systems.

Gradient Descent Risk Conditions

If learning rate is too high or function is not convex, it may miss the optimum.

Learning Rate in Gradient Descent

Controls how fast weights are updated; too high may overshoot, too low may be slow.

Overfitting with Small Datasets

Model may memorize training examples due to lack of generalization data.

Null Hypothesis in ANOVA

Group means are equal.

Alternative Hypothesis in ANOVA

At least one group mean is different.

SQL Definition

Structured Query Language used to manage and query relational databases.

Purpose of PCA

Reduce data dimensions while retaining most of the variance.

Ensemble Learning

Randomization helps reduce overfitting and improve model diversity.

Encoding Example

Gender as Male=1, Female=0 or one-hot: [1,0], [0,1].