MACHINE LEARNING FOR EMAIL
SPAM FILTERING (C951 TASK 3)
1. When designing a Naive Bayes classifier for spam detection, why is Laplace smoothing
(additive smoothing) essential?
A. To prevent the model from overfitting on rare words.
B. To reduce the computational complexity of the logarithm operations.
C. To handle the zero-frequency problem where an unseen word results in a zero posterior
probability.
D. To automatically filter out stopwords during the training process.
Answer: C
Conceptual Explanation: Laplace smoothing adds a small value to every count, ensuring
that probabilities are never zero, which would otherwise nullify the entire product in Naive
Bayes if a single feature is missing.
2. Which metric is generally considered most critical when minimizing the risk of ‘False
Positives’ in a production spam filter?
A. Recall
B. Precision
,C. F1-Score
D. Accuracy
Answer: B
Conceptual Explanation: Precision measures the ratio of correctly predicted positive
observations to the total predicted positives. In spam filtering, a high precision means
fewer legitimate emails (False Positives) are incorrectly flagged as spam.
3. How does TF-IDF (Term Frequency-Inverse Document Frequency) improve upon simple
Count Vectorization in spam filtering?
A. By increasing the weight of common words like ‘the’ and ‘and’.
B. By reducing the dimensionality of the feature space through principal component
analysis.
C. By penalizing words that appear frequently across all emails, highlighting unique spam-
indicative keywords.
D. By treating word order and context as primary features.
Answer: C
Conceptual Explanation: TF-IDF reduces the weight of terms that appear very frequently
across the entire corpus (IDF) while boosting terms that are frequent within specific
documents (TF).
, 4. In the context of the C951 proposal, what is ‘Concept Drift’ in email spam filtering?
A. The deterioration of model performance over time as spammer tactics evolve.
B. The process of moving the model from a development environment to a cloud
production environment.
C. The mathematical movement of data points across a Support Vector Machine
hyperplane.
D. The variation in computational resources required to process large email attachments.
Answer: A
Conceptual Explanation: Concept Drift occurs when the statistical properties of the target
variable (spam vs. ham) change over time, requiring model retraining.
5. Why might a Support Vector Machine (SVM) using a Radial Basis Function (RBF) kernel
outperform a Linear kernel for spam detection?
A. Linear kernels are incapable of handling high-dimensional text data.
B. RBF kernels are computationally faster for millions of features.
C. Linear kernels require all input data to be pre-normalized to a range of [0,1].
D. RBF can capture non-linear relationships and complex decision boundaries between
spam and ham.
Answer: D
SPAM FILTERING (C951 TASK 3)
1. When designing a Naive Bayes classifier for spam detection, why is Laplace smoothing
(additive smoothing) essential?
A. To prevent the model from overfitting on rare words.
B. To reduce the computational complexity of the logarithm operations.
C. To handle the zero-frequency problem where an unseen word results in a zero posterior
probability.
D. To automatically filter out stopwords during the training process.
Answer: C
Conceptual Explanation: Laplace smoothing adds a small value to every count, ensuring
that probabilities are never zero, which would otherwise nullify the entire product in Naive
Bayes if a single feature is missing.
2. Which metric is generally considered most critical when minimizing the risk of ‘False
Positives’ in a production spam filter?
A. Recall
B. Precision
,C. F1-Score
D. Accuracy
Answer: B
Conceptual Explanation: Precision measures the ratio of correctly predicted positive
observations to the total predicted positives. In spam filtering, a high precision means
fewer legitimate emails (False Positives) are incorrectly flagged as spam.
3. How does TF-IDF (Term Frequency-Inverse Document Frequency) improve upon simple
Count Vectorization in spam filtering?
A. By increasing the weight of common words like ‘the’ and ‘and’.
B. By reducing the dimensionality of the feature space through principal component
analysis.
C. By penalizing words that appear frequently across all emails, highlighting unique spam-
indicative keywords.
D. By treating word order and context as primary features.
Answer: C
Conceptual Explanation: TF-IDF reduces the weight of terms that appear very frequently
across the entire corpus (IDF) while boosting terms that are frequent within specific
documents (TF).
, 4. In the context of the C951 proposal, what is ‘Concept Drift’ in email spam filtering?
A. The deterioration of model performance over time as spammer tactics evolve.
B. The process of moving the model from a development environment to a cloud
production environment.
C. The mathematical movement of data points across a Support Vector Machine
hyperplane.
D. The variation in computational resources required to process large email attachments.
Answer: A
Conceptual Explanation: Concept Drift occurs when the statistical properties of the target
variable (spam vs. ham) change over time, requiring model retraining.
5. Why might a Support Vector Machine (SVM) using a Radial Basis Function (RBF) kernel
outperform a Linear kernel for spam detection?
A. Linear kernels are incapable of handling high-dimensional text data.
B. RBF kernels are computationally faster for millions of features.
C. Linear kernels require all input data to be pre-normalized to a range of [0,1].
D. RBF can capture non-linear relationships and complex decision boundaries between
spam and ham.
Answer: D
