Unit 4: AI with Orange Data Mining Tool

Accuracy measures the overall fraction of predictions that match the actual classes. For a binary classification confusion matrix, accuracy = (TP + TN) / (TP + TN + FP + FN), where TP and TN are correct decisions. Option B is precision, calculated as TP/(TP+FP); option C is recall, TP/(TP+FN); and option D is the F1 score, so A is the only correct definition.

In Orange Data Mining, the Data Table widget presents the selected dataset as a table, with records shown in rows and attributes or variables shown in columns. This makes it possible to inspect values, identify missing entries, and check whether the data were loaded as expected. File is used mainly to import data, Select Columns changes the selected variables, and Test & Score evaluates learners. Therefore, A is correct.

Responsible AI aims to develop and use systems that are safe, fair, transparent, accountable, and respectful of people’s rights. Appropriate human oversight is retained so that important decisions can be reviewed and corrected when necessary. Spreading misinformation, removing all control, and encouraging harmful bias are contrary to responsible AI principles.

Model evaluation measures performance on test data or other data that were not used to fit the model. It helps estimate generalization, reveal errors, and compare suitable performance measures such as accuracy, precision, recall, or mean error. A model can perform well on training examples but poorly on new cases because of overfitting. Evaluation therefore supports a safe deployment decision. Options B, C, and D are not purposes of evaluation, so A is correct.

Clustering groups data examples according to their similarities without using predefined target labels. It is therefore an unsupervised-learning method. Supervised learning uses labelled examples, while reinforcement learning uses rewards or feedback to guide actions.

A decision tree uses internal nodes to represent questions or conditions and branches to represent possible answers. Following a path from the root to a leaf produces a prediction or decision. Thus, option A correctly describes the method, while the other options contradict its data-based branching process.

The nearest-neighbor method compares a new data point with stored training examples using a distance or similarity measure. It bases the prediction on the closest examples; in classification, the majority class among those neighbors is commonly selected. Therefore, option A is correct.

Clustering is an unsupervised learning method that groups data points according to similarity or distance. Points with similar features are generally placed in the same cluster. It does not aim to isolate every point, and it does not require class labels provided in advance. Thus, similarity is central to forming meaningful clusters.

A false negative occurs when a person who actually has a disease receives a negative screening result. This may delay diagnosis, treatment, and preventive action. Option B describes a false positive, not a false negative. Sensitivity or recall is especially important when the aim is to miss as few genuine cases as possible.

Classification assigns an observation to one class from a defined set of labels. “Sick or not sick” has two discrete outcomes, so it is a binary-classification output and option A is correct. Estimated house price, next-day temperature, and sales amount are numerical quantities that are normally predicted using regression, not classification.

In supervised learning, each training input is paired with its correct output or label. The model learns the relationship between inputs and labels, such as classifying an email as “spam” or “not spam.” In contrast, unsupervised learning has no predefined labels and finds groups or patterns in the data. Therefore, option A is correct, while option B incorrectly describes unsupervised learning as using labeled data.

Clustering is an unsupervised learning method that forms groups of data points with similar characteristics without predefined class labels. Therefore, it is suitable for identifying groups of similar customers based on features such as purchasing behaviour. Option B refers to classification, where the correct class of each example is already known.

Recall is the proportion of actual positive cases, such as people who truly have a disease, that are correctly identified. In medical screening, classifying a diseased person as healthy is a false negative and can have serious consequences. Therefore, high recall may be prioritised to identify as many diseased cases as possible. Option B is incorrect because false positives should not be ignored; measures such as precision are used to reduce them.

Precision measures how many messages predicted as spam are actually spam. High precision means fewer false positives, reducing the chance that legitimate messages are incorrectly moved to the spam folder. Options B, C, and D are not purposes of precision. Exam tip: precision focuses on the correctness of predicted positive cases.

Sentiment analysis identifies the attitude or emotional polarity expressed in text. The phrases “very useful” and “made studying easier” communicate approval, benefit, and satisfaction, so the sentence has positive sentiment and option A is correct. It is not neutral because it contains an evaluation, not negative because no criticism is expressed, and not mixed because opposing emotions are absent.