Machine Learning & Predictive AI

Identify and handle missing values using imputation strategies appropriate to the data type.

Encode categorical variables using one-hot, label, and target encoding strategies.

Apply feature scaling (standardization, min-max normalization) and explain when each is needed.

Apply Principal Component Analysis (PCA) to reduce dimensionality while retaining variance.

Apply filter methods (e.g., univariate statistical tests) for feature selection.

Apply wrapper methods (e.g., recursive feature elimination) for feature selection.

Detect outliers using statistical methods such as Z-score and Interquartile Range (IQR).

Formulate strategies for handling outliers based on model requirements and domain knowledge.

Explain Maximum Likelihood Estimation (MLE) as a general principle for parameter fitting.

Derive and interpret the cost function for ordinary least squares (OLS) regression.

Implement gradient descent to optimize linear regression parameters.

Evaluate the assumptions of linear regression using residual diagnostics.

Explain the logistic regression model, including the sigmoid function and log-loss cost function.

Apply Bayes' theorem to explain the Naive Bayes classification algorithm and its conditional independence assumption.

Compute and interpret a confusion matrix, precision, recall, and F1-score.

Plot and interpret ROC curves and compute AUC to compare classifiers.

Fit polynomial regression models and assess the tradeoff between flexibility and overfitting.

Apply Ridge regression (L2 regularization) to reduce overfitting in high-variance models.

Apply Lasso regression (L1 regularization) for feature selection and model shrinkage.

Compare Elastic Net with Ridge and Lasso for balancing regularization strategies.

Classify data using Support Vector Machines with linear and kernel-based approaches.

Extend binary classification to multi-class problems using one-vs-rest (OvR) and softmax regression.

Apply techniques to handle class imbalance (e.g., SMOTE, class weights).

Explain decision tree splitting criteria (Gini impurity, information gain) and control complexity with pruning.

Explain how Random Forests reduce variance through bagging and feature randomness.

Describe the gradient boosting algorithm and its implementation in libraries like XGBoost or LightGBM.

Train a gradient boosting model for a regression or classification task.

Extract and interpret feature importance scores from tree-based ensemble models.

Perform hyperparameter tuning using grid search and randomized search.

Apply Bayesian optimization for more efficient hyperparameter tuning.

Distinguish between global and local model interpretability.

Apply model-agnostic interpretation techniques like Partial Dependence Plots (PDP) and Individual Conditional Expectation (ICE).

Calculate and visualize SHAP (SHapley Additive exPlanations) values to explain individual predictions.

Use SHAP summary plots to understand global feature importance and effects.

Perform systematic error analysis to identify model failure modes and areas for improvement.

Create technical documentation for a trained model, summarizing its performance, limitations, and assumptions.

Communicate model results and limitations effectively to technical and non-technical stakeholders.

Decompose a time series into trend, seasonal, and residual components.

Test for stationarity using statistical tests (e.g., Augmented Dickey-Fuller) and apply differencing to achieve it.

Interpret Autocorrelation (ACF) and Partial Autocorrelation (PACF) plots to identify model order.

Apply Holt-Winters exponential smoothing for series with trend and seasonality.

Specify, fit, and diagnose an ARIMA model using ACF/PACF and residual analysis.

Extend ARIMA to handle seasonal patterns using Seasonal ARIMA (SARIMA).

Implement walk-forward validation and evaluate forecast accuracy using metrics like MAE, RMSE, and MAPE.

Frame a time series forecasting problem as a supervised learning regression task.

Engineer lag features, rolling window statistics, and calendar features from a time series dataset.

Train and evaluate tree-based models (Random Forest, XGBoost) for time series forecasting.

Apply Facebook Prophet for forecasting with automatic trend and seasonality detection.

Compare single-step vs. multi-step forecasting strategies and their tradeoffs.

Implement a recursive forecasting strategy for multi-step predictions using ML models.

Describe the architecture of a Recurrent Neural Network (RNN) and its use of a hidden state for sequences.

Explain the architecture of LSTM and GRU networks and their advantages for sequential data.

Prepare time series data for a deep learning model by creating sequences of inputs and outputs.

Build and train a simple LSTM model for single-step time series forecasting.

Implement an encoder-decoder architecture for multi-step time series forecasting.

Evaluate and compare statistical, ML, and deep learning forecasting approaches and justify the choice for a given business scenario.

Serialize and version trained models using joblib/pickle and a model registry.

Design a REST API contract for serving predictions, including endpoints, request/response schemas, and error handling.

Implement the REST API using a web framework (e.g., FastAPI) to create a custom prediction service.

Containerize the prediction service for deployment using Docker.

Write unit and integration tests for the prediction service.

Deploy a containerized model as an online prediction endpoint on Google Vertex AI.

Configure and execute batch prediction jobs on Vertex AI for large-scale inference.

Design a CI/CD workflow for a machine learning model, outlining triggers, stages, and artifacts.

Implement a CI/CD pipeline to automatically build, test, and deploy a model to a Vertex AI endpoint.

Explain the principles of A/B testing and configure an A/B test on Vertex AI to evaluate competing models.

Distinguish between data drift and concept drift and explain their impact on model performance.

Select and implement statistical tests and metrics to detect data and concept drift.

Implement a feedback loop to capture prediction outcomes and compare against ground truth.

Design an automated retraining strategy, including trigger conditions, data sources, and model promotion criteria.

Implement an automated pipeline that retrains, evaluates, and versions a model upon drift detection.

Define and monitor SLAs, latency budgets, and performance benchmarks for production ML systems.

Architect an end-to-end ML system for a forecasting use case, creating a system diagram covering data ingestion, training, serving, and monitoring.