HomeSubjectsUniversityBlogAbout

AI / Machine Learning & Data Analytics

Concepts-first: you need to recognize models, not train them.

3,570+ MCQs17 topicsWeightage: 10%3 difficulty levels
Start Practice QuizBrowse 17 topics

Overview

AI / Machine Learning & Data Analytics is the subject where students with project experience tend to over-prepare and under-score. The NSCT section weights conceptual fluency — understanding what a model does, when to use it, what its failure modes are — over mathematical derivation. You will see questions on supervised vs unsupervised learning, linear and logistic regression, decision trees, overfitting, basic NumPy/Pandas operations, and how to read a confusion matrix. Deep learning shows up at a vocabulary level (forward pass, backprop, activation functions), not at the PyTorch line level.

Why This Subject Matters

AI/ML is a growing share of every Pakistani software house's hiring rubric. Even non-ML roles now expect a working vocabulary, and the NSCT section mirrors what recruiters quiz during interviews. Strong scores here translate directly into interview invitations.

Topics in AI / Machine Learning & Data Analytics

Foundations

Getting data into shape before modelling: loading, cleaning, missing-value strategies, and the probability/statistics baseline you are assumed to know.

Introduction to AI, ML & Data AnalyticsMathematical FoundationsPython for AIData Collection & Pre-processingExploratory Data Analysis

Classical ML

The bulk of the section: linear/logistic regression, decision trees, random forests, boosting, k-means, PCA, and how to measure each one's performance honestly.

Supervised LearningUnsupervised LearningEnsemble LearningFeature EngineeringModel Evaluation & Validation

Deep learning & applications

Vocabulary-level questions on neural networks, NLP basics, computer vision pipelines, big-data workflows, deployment patterns, and the ethics/security layer you should never skip.

Deep Learning FundamentalsAdvanced Deep LearningNLPComputer VisionBig Data AnalyticsModel Deployment & MLOpsAI Ethics, Security & Privacy

How to Study This Subject

Start with the model-selection decision tree in your head: given a problem type (classification, regression, clustering, dimensionality reduction), what are the two or three go-to models and when does each win? Once that map is solid, pivot to evaluation metrics (precision, recall, F1, AUC, MSE vs MAE) and the overfitting toolkit (regularization, cross-validation, train/test splits). Leave deep learning for last; it is a small slice of the marks.

Suggested time budget

Budget 12–15 hours. Spend roughly 60% on classical ML, 25% on evaluation metrics, and 15% on deep learning vocabulary.

Common Mistakes to Avoid

  • 1Memorizing formulas without understanding what each term represents or when it breaks.
  • 2Confusing precision and recall — and therefore choosing the wrong metric for imbalanced datasets.
  • 3Assuming more features always helps. The curse of dimensionality is a favourite NSCT trap.
  • 4Treating deep learning as a separate universe. Most of its NSCT questions boil down to basic ML concepts plus a layer name.

Sample Questions

Two example MCQs from the AI / Machine Learning & Data Analytics question bank, with full explanations. The live quiz draws from 3,570+ verified questions across three difficulty levels.

Model Evaluation & ValidationEasy

Q1. You train a model that gets 99% training accuracy but 62% test accuracy. What is most likely happening, and what is the first fix to try?

  1. A.Underfitting — train a larger model.
  2. B.Overfitting — add regularization or more data.✓ Correct
  3. C.Data leakage — the test set is too easy.
  4. D.The metric is wrong — switch to precision/recall.

Explanation

A large gap between training and test accuracy is the textbook signature of overfitting. The first line of defence is regularization (L1/L2, dropout) or increasing the training data. Underfitting would show low accuracy on both sets. Data leakage usually inflates test accuracy, not the reverse.

Model Evaluation & ValidationMedium

Q2. In a fraud detection dataset where 1% of transactions are fraudulent, which metric is MOST misleading if used alone?

  1. A.F1 score
  2. B.Precision
  3. C.Recall
  4. D.Accuracy✓ Correct

Explanation

A model that predicts 'not fraud' for every transaction scores 99% accuracy on this dataset while catching zero fraud. Accuracy hides class imbalance; precision and recall (and their harmonic mean, F1) reveal it.

Related NSCT Subjects

Databases

3,570+ MCQs

Programming

3,360+ MCQs

Ready to practice AI / Machine Learning & Data Analytics?

Choose your topic, pick a difficulty, and start answering. No signup required — your progress is saved in your browser.

Start AI / Machine Learning & Data Analytics Quiz