Unit 2: Data Science Methodology

Correct answer: A. When the classes are highly imbalanced

Accuracy can be misleading when classes are highly imbalanced. A model that predicts only the majority class may obtain a high accuracy while performing poorly on the minority class. In such cases, precision, recall, F1 score, and the confusion matrix provide additional information about classification performance.

Correct answer: A. To evaluate the balance between precision and recall

The F1 score is the harmonic mean of precision and recall, so it combines both measures into one value. It is useful when both false positives and false negatives matter, especially in classification tasks with imbalanced classes. It measures predictive classification performance, not execution time or file properties.

Correct answer: A. Very good performance on training data but weak performance on new data

Overfitting occurs when a model learns the training data too specifically and may partly memorize it. Consequently, it performs very well on training data but poorly on unseen or test data. Option B suggests better generalization, whereas options C and D describe failures or missing inputs rather than overfitting.

Correct answer: A. The ability to perform well on new, unseen data

Generalization is the ability of a machine-learning model to perform well on new data that was not used during training. It shows that the model has learned useful patterns rather than merely memorizing examples. Repeated training, memorization, or adding features alone does not define generalization.

Correct answer: A. To maintain a systematic record of the process, decisions, and results

Documentation records the project stages, decisions made, data used, and results obtained. This allows the team to review, reproduce, and improve the work when needed. Therefore, option A is correct; option D is incorrect because proper documentation supports model evaluation and improvement rather than preventing it. Exam tip: In the AI project cycle, associate documentation with transparency, reproducibility, and continuous improvement.

Correct answer: A. Identifying spam using labelled emails

Supervised learning trains a model with input examples paired with known labels. In option A, emails are labelled “spam” or “not spam,” so the model learns a classification mapping and can label new messages. Option B is unsupervised clustering, option C is reinforcement learning based on rewards, and option D is manual programming rather than machine learning.

Correct answer: A. Data acquisition

Problem scoping identifies the problem, goal, and stakeholders. The next stage is data acquisition, in which the information or data needed for the project is collected. Final deployment occurs later, while hiding results and deleting files are not recognised stages of the AI project cycle. Exam tip: remember the sequence as problem scoping → data acquisition → data exploration → modelling → evaluation.

Correct answer: A. The project direction and priorities may become unclear.

In problem scoping, a clear goal identifies the problem to be addressed, the work to prioritise, and the criteria for measuring success. If the goal is unclear, the team may not have a clear project direction or set of priorities; therefore, option A is correct. Option B is incorrect because an unclear goal does not narrow the scope automatically; it can instead make the scope uncertain or overly broad. A clear goal also does not remove the need for data collection or discussion with stakeholders.

Correct answer: A. Whether the solution is achieving the project’s defined objectives and expected outcomes

Success indicators are predefined, measurable criteria used to determine whether a solution is achieving the project’s objectives and expected outcomes. The amount of data collected or the time spent by a team may be project statistics, but they are not by themselves the primary measure of success unless they have specifically been set as project objectives.

Correct answer: A. Collecting individuals’ data with their informed consent

Option A is correct because informed consent means that individuals are told what data will be collected, why it will be collected, and how it will be used before agreeing. Option B violates privacy. Option C lacks transparency about the data source, while option D harms the accuracy and reliability of the dataset.

Correct answer: A. The source is authoritative, its information is verified, and it is up to date

A reliable data source has identifiable authority, contains verified information, and is current or regularly updated. Therefore, option A is correct. In contrast, the absence of author or organization details and publication or update dates makes it difficult to assess a source's accountability and currency.

Correct answer: A. The model may become biased toward the majority class

In imbalanced data, one class has many more examples than the other class or classes. The model may therefore learn the majority class more strongly, become biased toward it, and identify the minority class less accurately. Option B is incorrect because class imbalance does not automatically ensure fair performance for every class.

Correct answer: A. They can affect analysis results and model training.

Missing values indicate that some information is unavailable. If they are not handled, they can bias analysis results or prevent some models from being trained. Depending on the data and context, they may be removed, filled with suitable values, or marked as a separate category. Option B is incorrect because not all machine-learning algorithms handle missing values automatically and correctly.

Correct answer: A. To convert data into a suitable form for analysis or modeling

Data transformation changes raw data into a form that is suitable for analysis or use by a machine learning model. For example, it may involve scaling numerical values, encoding categories, or changing a data format. Therefore, option A is correct. Option B refers to deleting data, whereas transformation changes the data’s representation or format for use.

Correct answer: A. To select relevant features and reduce unnecessary dimensions

Feature selection chooses the most relevant and useful features for a model from the available features. It can remove irrelevant or redundant features, reduce dimensionality, and make the model simpler and more effective. Option B is incorrect because including all features without evaluation is not feature selection.

Correct answer: A. Estimated price of a house

Regression predicts a numerical value, often on a continuous scale. An estimated house price is expressed as an amount of money and can take many possible values, so option A is the suitable output. Cat versus dog, spam versus not spam, and red versus yellow are choices among labels; these are classification outputs rather than regression outputs.

Correct answer: A. Overfitting

This indicates overfitting. The model learns the training data too closely, including noise or accidental patterns. Therefore, it achieves high accuracy on training data but performs poorly on new test data. With good generalization, the gap between training and test performance is usually not very large.

Correct answer: A. Underfitting

In underfitting, the model fails to learn the patterns in the training data adequately, so its performance is poor on both the training and testing data. In overfitting, training performance is usually good while testing performance is poor. Exam tip: Poor performance on both datasets suggests underfitting, whereas poor performance mainly on testing data suggests overfitting.

Correct answer: A. The model's positive prediction was correct

In a confusion matrix, a true positive occurs when the actual condition is positive and the model also predicts positive. Thus the prediction is both positive and correct, making option A the answer. Option B is a false positive because a negative case was labelled positive; option C is a false negative, and option D describes a true negative.

Correct answer: A. The actual class is negative, but the model predicts it as positive.

A false positive occurs when the actual class is negative, but the model labels it as positive. Therefore, option A is correct. For example, incorrectly identifying a person without a disease as disease-positive is a false positive and may lead to unnecessary follow-up testing. Option B is a true positive, whereas option D represents a false negative.

Correct answer: A. To detect changes in data, performance, and errors during real-world use

After deployment, a model works on new, real-world data. Over time, data patterns, user behaviour, or model performance may change, which can increase errors or reduce accuracy. Monitoring helps identify such problems in time so that the model can be improved or updated when needed. Unlike option C, evaluating a model only on training data does not show how it performs in real-world use.

Correct answer: A. To clearly present the project process, findings, and learning

A final report systematically records the problem, data used, method followed, results, limitations, and learning. Therefore, option A is correct because it presents the project process and findings transparently. Option D is incorrect because results must be evaluated and checked before they are reported. Exam tip: Remember that a final report supports documentation, communication, and future improvement of the project.