question correct answer explanation

Which of the following is a f1 score The Fl score is a common metric for evaluating the performance of a
method for evaluating the binary classification model, which takes into account both precision
performance of a binary and recall.
classification model?
Which ML type is used in Unsupervised Clustering is a classic unsupervised learning technique used to group
clustering similar customers in Learning similar data points like customer segmentation.
marketing?
In decision trees, Limit the The 'max_features' parameter limits the number of features that are
'max_features' parameter is number of considered for making splits at each node, providing a way to
used to: features control overfitting.
considered for
splitting at
each node
A decision tree with only one Stump A decision tree with only one node (the root) is called a stump.
node (the root) is called: It represents the simplest form of a decision tree.
What does a silhouette score High intra- A silhouette score close to +1 indicates that the data points
value close to +1 indicate? cluster within each cluster are similar to one another (high intra-
similarity and cluster similarity) and dissimilar to the data points in other
low inter- clusters (low inter-cluster similarity). It suggests a well-
cluster separated and cohesive cluster structure.
similarity
Which of the following is a Confusion
method for evaluating the matrix
performance of a multi-class
classification model?
Which of the following is a Both a & b
technique for handling missing
data in a supervised learning
model?
What are the three main Data cleaning, The three main stages of data processing in machine learning are
stages of data processing in preprocessing, data cleaning, preprocessing, and feature engineering. These stages
machine learning? and feature involve transforming raw data into a format that can be used to train
engineering a machine learning model.
What is the primary purpose Image CNNs are specifically designed for processing data that has a grid-
of a Convolutional Neural Recognition like topology, such as images.
Network (CNN}?
Which of the following is a sign Performs High bias typically leads to underperformance on both training
of a high bias model? poorly on both and test datasets, as the model is too simplistic to capture the
training and underlying patterns of the data.
test data
Which of the following is a All of the Removing the outliers from the dataset, winsorizing the data, and
technique for reducing the above using a robust loss function are all techniques for reducing the
impact of outliers on a impact of outliers on a supervised learning model.
supervised learning model?
In k-means clustering, 'k' Number of In k-means clustering, 'k' represents the number of clusters
stands for: clusters into which the data is to be grouped.
What is the purpose of using a To prevent A train-test split in linear regression is used to evaluate how
train-test split in linear overfitting and well the model performs on unseen data. It helps in preventing
regression modeling? assess the overfitting and in assessing the model's ability to generalize to
model's new data.
generalization
capability
In linear regression, what role It indicates the R-squared is a statistical measure that represents the
does the R-squared (R2 } proportion of proportion of the variance for the dependent variable that's
statistic play? variance in the explained by the independent variables in the model.
 dependent
variable
explained by
the model
Statement: Clustering Clustering The statement accurately describes the purpose of clustering
algorithms are used to group algorithms find algorithms. Clustering techniques, such as K-Means,
similar data points together patterns and DBSCAN, and Hierarchical Clustering, are used to discover
based on certain criteria. structure in inherent patterns and groupings within unlabeled data. These
Which of the following options unlabeled algorithms partition the data into clusters based on similarity
correctly explains the data. or density, helping to identify relationships and uncover
statement? underlying structure.
What is the main difference Supervised Supervised learning requires labeled data, where the system
between supervised and learning is trained on input-output pairs. Unsupervised learning, on the
unsupervised learning? requires other hand, does not require labeled data and focuses on
labeled data, finding patterns in the data.
while
unsupervised
learning does
not
Which of the following is true Lowering bias This trade-off implies that decreasing bias might increase
about the bias-variance trade- may increase variance and decreasing variance might increase bias. A
off? variance, and balance is needed for optimal model performance.
vice versa
Which of the following is an Clustering Clustering users into groups without prior labels is a classic
example of unsupervised users based unsupervised learning task.
learning? on browsing
patterns
What does the slope of the The expected In simple linear regression, the slope indicates how much the
regression line represent in change in the dependent variable is expected to increase (or decrease)
simple linear regression? dependent when the independent variable increases by one unit.
variable for a
one-unit
change in the
independent
variable
Which clustering algorithm DBSCAN Density-Based Spatial Clustering of Applications with Noise
does not require specifying the (DBSCAN} does not require specifying the number of clusters
number of clusters beforehand. It groups data points based on their density and
beforehand? identifies clusters as regions of high-density separated by
regions of low-density. DBSCAN can handle clusters of
arbitrary shape and is effective in identifying outliers as noise
points.
What is the main goal of To enable Machine learning enables systems to learn from data and make
Machine Learning? machines to decisions or predictions without being explicitly programmed.
learn patterns
and make
decisions from
data
Which of the following is true It involves Unsupervised learning involves finding patterns in data
about unsupervised learning? finding without specific labels or target outputs. Unlike supervised
patterns in learning, it does not require labeled data for training.
data without
specific labels
What does high bias in a The model is High bias typically indicates that the model is too simple and
machine learning model underfitting does not capture the complexity of the data, leading to
indicate? underfitting. It often results in poor performance on both the
training and validation datasets.
A decision tree that has not Overfit the An unpruned decision tree, especially if it is allowed to grow to
been pruned is likely to: training data its full depth, is likely to overfit the training data, capturing
noise and anomalies in the training data.
Decision trees can handle: Both numerical Decision trees are capable of handling both numerical and
and categorical categorical data, making them versatile for various types of
data datasets.
Can decision trees be used for Yes, but not While decision trees can technically be used for time-series
time-series analysis? effectively analysis, they are generally not effective for this purpose due
to their inability to capture temporal dependencies.
 What is recall? The number of
true positives
divided by the
sum of true
positives and
false negatives
What is a disadvantage of They are prone One of the main disadvantages of decision trees is their
using a decision tree? to overfitting tendency to overfit, especially if they areallowed to grow
without constraints like depth limitation or pruning.
Which of the following is NOT Distributed Sorry,you are incorrect !Distributed learning is a technique,
a main type of machine Learning not one of the three main types of ML (supervised,
learning? unsupervised, reinforcement).
K-means clustering is an False K-means clustering is not a hierarchical clustering algorithm.
example of a hierarchical It is a partition-based clustering algorithm that aims to divide
clustering algorithm. data points into K clusters, where K is a predetermined
number.
Which metric is typically not Mean Squared Mean Squared Error is typically used for regression tasks, not
used to evaluate the Error for classification models like logistic regression.
performance of a logistic
regression model?
What is a decision tree mainly Data Decision trees are primarily used for classification tasks in
used for in machine learning? Classification machine learning, although they can also be used for
regression.
What does the silhouette score Cluster The silhouette score measures how similar a data point is to
measure in clustering? separation its own cluster compared to other clusters, indicating the
separation between clusters.
Which of these problems is Forecasting Sales forecasting is a supervised task where historical labeled
best solved using supervised next week's data is used to predict future values.
learning? sales
What is underfitting? When a model Underfitting occurs when a machine learning model is too
is too simple simple and fails to capture the underlying patterns in the data.
and fails to This can lead to the model performing poorly on both the
capture the training data and new, unseen data.
underlying
patterns in the
data
Which function is used in Sigmoid Logistic regression uses the sigmoid function to map predicted
logistic regression to map function values to probabilities.
predicted values to
probabilities?
Which type of ML is used when Unsupervised Unsupervised learning deals with unlabeled data to find
data has no labels? Learning hidden patterns or structures.
Which supervised learning Logistic Logistic regression is a commonly used supervised learning
algorithm is commonly used for regression algorithm for classification problems. It is a variant of linear
binaRY classification regression that models the probability of the target variable
problems? belonging to a particular class.
Decision trees are particularly There is a One of the key advantages of decision trees is their
useful when: need for model interpretability. They provide a clear visualization of the
interpretability decision-making process, making them useful in scenarios
where understanding the model's decisions is important.
How does Lasso regression By setting the Lasso regression performs feature selection by applying Ll
perform feature selection? coefficients of regularization, which can shrink some coefficients to exactly
less important zero. This effectively removes those features from the model.
features to
zero

Forecasting and Budgeting:

1. Walk me through the budgeting process you managed in your previous role.
Answer: The budgeting process at I-DAPT HUB Foundation was heavily project-
and grant-driven, requiring a blend of operational forecasting and strict
compliance with government grant guidelines (like those under NM-ICPS). I
collaborated with the CEO, the Project Director, and the heads of the five Thrust
Areas to define the upcoming year's key initiative and then I created a Budget
 outlining the anticipated grant disbursements for new projects. I led the
reconciliation, challenging budget owners on key cost drivers. For instance, I
might question a high consultancy budget if internal staff had a clear mandate,
ensuring we optimized the use of restricted grant funds versus flexible
institutional funds. I prepared the first draft financial package for the Board,
including the Statement of Activities (P&L), Cash Flow Forecast, and a
Grant Utilization Schedule for restricted funds. I presented the final budget to
the Board of Directors, focusing the discussion on strategic reserves, the
financial viability of self-sustaining initiatives (like STCs), and the burn rate
against the multi-year NM-ICPS grant.

2. What are the different forecasting methodologies (e.g., driver-based, rolling

forecast, zero-based budgeting), and when would you use each?
Answer: The selection of a forecasting methodology in FP&A depends critically on
the organization's needs: Driver-Based Forecasting should be utilized for
strategic planning in complex models, as it links financials (like revenue or
COGS) to key operational metrics (drivers), enabling accurate 'what-if' scenario
modelling and understanding of operational impact. Conversely, the Rolling
Forecast provides continuous, near-term agility by constantly updating the
forecast horizon (e.g., 12-18 months), making it ideal for dynamic or volatile
markets where rapid response to changes is paramount. Finally, Zero-Based
Budgeting (ZBB) is typically reserved for instances requiring fundamental cost
control and efficiency improvements, as it forces the justification of every
expense from a zero base, ensuring all spending aligns with current strategic
priorities, often during major restructuring or cost-challenging initiatives.

3. How do you handle significant variance analysis (Budget vs. Actuals)? Describe
a time when you identified a major variance and what action you recommended.

Answer: My approach to significant variance analysis is a three-step process:

Calculation, Investigation, and Recommendation. At I-DAPT, I identified a
15% unfavourable variance in the Defence R&D Thrust Area's Equipment
Consumables within a restricted NM-ICPS grant. The investigation revealed a mix
of an uncontrollable 20% Price Variance on a specialized server component
(due to inflation) and a deliberate Timing Variance (early Q4 purchase pulled
into Q3). I formally proposed an internal fund re-allocation from a favorable
Travel budget to cover the price increase, ensuring grant compliance, and
immediately revised the Q4 rolling forecast downwards to reflect the pulled-
forward timing.
4. Describe a complex financial model you built at I- DAPT HUB foundation. What
were the key drivers and assumptions?

Answer: The most complex model I built was the 5-Year Sustainability and
Cash Flow Model for the NM-ICPS program review, designed to project the Hub's
financial self-sufficiency post-grant. The complexity lay in integrating five
distinct revenue streams into a single forecast. Key drivers included the
Enrolment Rate and Course Frequency for training revenue, and the Project
Conversion Rate for consulting. Critical assumptions included a 10% annual
enrolment growth rate, a conservative 7% annual salary increase, and a
scheduled 5-year Capex refresh cycle.

Here are the expected follow-up questions for the statement: "The model was
instrumental, proving that a dedicated focus on monetizing high-volume Training
(STCs) was required to build reserves and reach the breakeven point, effectively
guiding the leadership’s focus on achieving long-term operational viability."
 Expected Follow-up Questions
1. "Can you elaborate on how the model proved that Training (STCs) was
the key to reaching the breakeven point? What specific metrics did you
present to leadership?" (Focuses on evidence and presentation skills)
2. "What operational or financial changes did leadership implement based
on your recommendation to focus on STCs? Can you quantify the positive
impact of those changes?" (Focuses on impact, execution, and quantifiable
results)
3. "Since the focus shifted to STCs, what were the major risks or trade-offs
you identified regarding resource allocation, especially concerning core
research activities?" (Focuses on strategic thinking and risk management)
4. "How did you monitor the progress of the STC monetization strategy
against your model's projections? What was your process for
implementing a forecast adjustment when the actual results deviated?"
(Focuses on monitoring, variance analysis, and the rolling forecast process)
5. "You mentioned 'long-term operational viability.' What was the timeline
or horizon for achieving this, according to your model, and what was the
defined 'breakeven point' metric?" (Focuses on definition of success and key
metrics)

Answer: The model was instrumental, proving that a dedicated focus on

monetizing high-volume Training (STCs) was required to build reserves and reach
the breakeven point, effectively guiding the leadership’s focus on achieving long-
term operational viability.