Roll#:

Course: Introduction to Data Science Course Code: DS 2001

Program: BS(DS) Semester: Fall 2022
Duration: 3 Hour Total Marks: 90
Paper Date: 27-12-2023 Page(s): 12
Section: BS (DS) A, B, C Section:
Exam: Final Roll No:
Instructions: Answer in the space provided. You can ask for rough sheets, but they will not
be graded or marked. In case of confusion or ambiguity make a reasonable
assumption. Questions during exam are not allowed.

Question#1 [6+10+5=21]

Given the following dataset

x1 x2 y Predicted y

1 2 1

-2 -2 0

1 -1 0

3 1 1

a) Draw the architecture of a Multilayer Perceptron network containing 2 neurons and an

output layer with a single neuron. Your network should take input features x1 and x2.
b) Using the architecture from Part a, perform a single forward pass with the given
specifications:
i) All weights for the input layer are set to 0.1 and Bias is set to 1.
ii) Output layer weights are set to 0.2 with bias of 0.5.
iii) Apply the binary step activation function to each neuron
c) Using a threshold of >=0.5, convert the output of the output neuron to 0 and 1. Compare the
predicted output with the actual output provided and calculate the accuracy of the model.

1
Roll#:

x x2 y Hidden Layer Output Layer Predicted y

1 T>=0.5

1 2 1 a = 1*0.1+2*0.1 + 1 = 1.3 c = 0.78*0.2 + 0.78*0.2 + 0.5 = 0.81 1

Activation = Sigmoid
S(a) = 0.78
b = 1*0.1+2*0.1 +1 = 1.3
S(a) = 0.78

-2 -2 0 a = -2*0.1+(-2)*0.1 + 1 = 0.6 c = 0.64*0.2 + 0.65*0.2 + 0.5 = 0.75 1

S(a) = 0.64
b = -2*0.1+(-2)*0.1 + 1 = 0.6
S(a) = 0.64

1 -1 0 a = 1*0.1+(-1)*0.1 + 1 = 1 c = 0.73*0.2 + 0.73*0.2 + 0.5 = 0.79 1

S(a) = 0.73
b = 1*0.1+(-1)*0.1 +1 = 1
S(a) = 0.73

3 1 1 a = 3*0.1+1*0.1 + 1 = 1.4 c = 0.8*0.2 + 0.8*0.2 + 0.5 = 0.82 1

S(a) = 0.8
b = 3*0.1+1*0.1 +1 = 1.4
S(a) = 0.8

Accuracy = 2/4 = 50%

2
Roll#:

3
Roll#:

Question#2: [3x6 = 18]

A financial institution has implemented a new fraud detection system to identify and prevent
fraudulent transactions. The system uses a machine learning model to analyze transaction patterns
and flag potentially fraudulent activities. Company is primarily interested in the number of
fraudulent transactions that are correctly identified by the machine learning model. The institution
wants to assess the performance of the system and has provided a dataset containing labeled
examples of transactions (fraudulent or legitimate) for this purpose.
Dataset:
The dataset consists of 501 transactions, with 351 legitimate transactions and 150 fraudulent
transactions. Each transaction record includes various features such as transaction amount,
transaction type, and time of day. However, the model fails to correctly identify 6 legitimate and 15
fraudulent transactions.
a) Create a confusion matrix for the fraud detection system. Clearly mention the predicted and
actual columns.
b) Calculate Accuracy, Precision, Sensitivity, Specificity and the F1 score.
c) Briefly interpret your results.

Acual

legitimate fraudulent
Predicted
legitimate 345 15

fraudulent 6 135

True Positive (TP): 150 - 15 = 135

True Negative (TN): 351 - 6 = 345
False Positive (FP): 6
False Negative (FN): 15

Accuracy = 95.8
Precision = 95.74
Recall/Sensitivity = 90
Specificity = 98.29
F1 Score = 0.927

4
Roll#:

Question#3 [10]

A marketing department is analyzing the effectiveness of their advertising campaigns across

different channels. They have collected data on the amount spent on advertising (in thousands) and
the corresponding sales revenue generated (in thousands) for a set of products. The marketing team
is interested in understanding the correlation between the advertising spending and sales revenue
to optimize their future campaigns. The dataset contains data for 20 products, with the advertising
spending and sales revenue for each product.
Create the heat map/correlation matrix for the above scenario, and comment on the importance of
each of the features in ML training.
Product Advertising Spending (pkr) Sales Revenue (pkr) Customer Satisfaction
Score (1-10)

1 100 500 7
2 150 470 5
3 200 700 8
4 120 300 4
5 180 800 9
6 250 400 6
7 90 600 7
8 300 250 3
9 170 900 8
10 130 520 6
11 160 630 7
12 220 550 5
13 110 480 4
14 190 730 9
15 140 720 8
16 260 200 2
17 230 950 10
18 120 380 5
19 200 830 9
20 180 700 8

5
Roll#:

Advertising Spending (pkr) vs. Sales Revenue (pkr):

Correlation Coefficient: -0.084
Interpretation: There is a weak negative correlation between Advertising Spending and Sales
Revenue. This suggests that as Advertising Spending increases, Sales Revenue tends to slightly
decrease. However, the correlation is not strong.

Advertising Spending (pkr) vs. Customer Satisfaction Score (1-10):

Correlation Coefficient: -0.116
Interpretation: There is a weak negative correlation between Advertising Spending and Customer
Satisfaction Score. This suggests that as Advertising Spending increases, Customer Satisfaction
Score tends to slightly decrease. Again, the correlation is not strong.

Sales Revenue (pkr) vs. Customer Satisfaction Score (1-10):

Correlation Coefficient: 0.931
Interpretation: There is a strong positive correlation between Sales Revenue and Customer
Satisfaction Score. This suggests that as Sales Revenue increases, Customer Satisfaction Score tends
to significantly increase. This strong positive correlation indicates a potential relationship between
the two variables.

6
Roll#:

Question#4: [10]

In this dataset:

● "Age" is the age of the customer.

● "Income" is the annual income of the customer.
● "Purchase" is the target variable indicating whether the customer made a purchase (Yes) or
not (No).

Calculate entropy and information gain [For section B and C]

id age Income purchase
1 20 - 25 low no
2 30 - 35 medium no
3 20 - 25 high yes
4 20 - 25 low no
5 20 - 25 high yes
6 30 - 35 low yes
7 20 - 25 high yes
8 30 - 35 low no
9 30 - 35 medium yes
10 20 - 25 medium no

OR

You are given a set of sentences, and your task is to calculate the Term Frequency-Inverse Document
Frequency (TF-IDF) for each word in these sentences. [For section A]
1. Inflation has increased unemployment
2. The company has increased its sales
3. Fear increased his pulse

7
Roll#:

8
Roll#:

9
Roll#:

Question#5 [10+5=15]

In this scenario, you are provided with data on several locations where accidents occurred in the
past month, along with the coordinates of three hospitals. Your objective is to analyze accident
hotspots and strategically cluster accident locations to determine an optimal allocation of hospitals.

Point

Name Latitude Longitude

1 31.5764 74.3118

2 31.5769 74.3131

3 31.5774 74.3149

4 31.5775 74.3177

5 31.5779 74.3194

6 31.5763 74.3173

7 31.5759 74.3199

8 31.5745 74.3175

9 31.5730 74.3199

Malik Jalal (74.3140, 31.5788) 10 31.5716 74.3169

Lady Aitchison ( 74.3155, 31.5739) 11 31.5747 74.3138
Mayo (74.3150,31.5722)
12 31.5736 74.3128
(a) Utilize clustering techniques to identify 2 accident
hotspots. Implement a clustering algorithm (e.g., K-Means) to group accident locations
based on proximity. Take 1 and 9 as initial centroids. Distance metric is Manhattan distance
(|(x2-x1)+(y2-y1)|).
(b) Determine the optimal allocation of hospitals to the identified accident hotspots. Assign
each cluster to the nearest hospital. Mark the resultant clusters on the map.

10
Roll#:

C1 = {1,2,3,11,12}
C2 = {4,5,6,7,8,9,10}

New centroids = [31.57580,74.31328],[31.57524,74.318271]

C1 = {1,2,3,11,12}
C2 = {4,5,6,7,8,9,10}

No change

11
Roll#:

12
Roll#:

Question#6 [2+2+2+10 = 16]

Suppose you are hired to analyze the dynamic relationship between the film industry and
socio-economic factors such as crime rate, GDP etc. in the United Kingdom from 2000 to 2022.
You are provided with three CSV files containing data related to IMDb movies and the crime rates.
The datasets and their attributes are as follows.

1- imdb_movies.csv:
Attributes: {Title, Genre, Rating, IMDb Rating, Year, Duration, RegionCode}
2- imdb_regions.csv:
Attributes: {RegionCode, Country, Language}
3- crime_rate.csv:
Attributes: {Year, Crime rate, Country}

a) Load the csv files into pandas dataframes and perform basic preprocessing to clean the
dataset.

# Load CSV files into DataFrames

imdb_movies = pd.read_csv('imdb_movies.csv')
imdb_regions = pd.read_csv('imdb_regions.csv')
crime_rate = pd.read_csv('crime_rate.csv')

#drop duplicates
imdb_movies .drop_duplicates(inplace=True)
imdb_regions .drop_duplicates(inplace=True)
crime_rate .drop_duplicates(inplace=True)

# Handle missing values

imdb_movies .dropna(inplace=True)
imdb_regions .dropna(inplace=True)
crime_rate .dropna(inplace=True)
# Convert data types if needed

crime_rate ['Crime rate'] = crime_rate ['Crime rate'].astype(float)

b) Merge the dataframes where necessary and filter the data that belongs to United Kingdom
during 2000-2022

merged_data = pd.merge(imdb_movies , imdb_regions , on='RegionCode', how='left')

merged_data = merged_data[(merged_data['Country'] == 'United Kingdom') &

(merged_data['Year'].between(2000, 2022))]

13
Roll#:

crime_rate = crime_rate [(crime_rate ['Country'] == 'United Kingdom') & (crime_rate

['Year'].between(2000, 2022))]

c) Perform EDA and come up with at least two comprehensive graphs/visualizations that can
help understand the data.

# Time Series Plot for Crime Rate Over the Years (2000-2022)
plt.figure(figsize=(12, 6))
sns.lineplot(x='Year', y='Crime rate', hue='RegionCode', data=crime_rate)
plt.title('Crime Rate Over the Years (2000-2022) in the United Kingdom')
plt.xlabel('Year')
plt.ylabel('Crime Rate')
plt.legend(title='Region Code', bbox_to_anchor=(1.05, 1), loc='upper left')
plt.show()

# Distribution of Movie Genres in the United Kingdom

plt.figure(figsize=(12, 6))
sns.countplot(x='Genre', data=merged_data, order=merged_data['Genre'].value_counts().index)
plt.title('Distribution of Movie Genres in the United Kingdom (2000-2022)')
plt.xlabel('Movie Genre')
plt.ylabel('Count')
plt.xticks(rotation=45, ha='right')
plt.show()

d) Your task is to find any relationship between movie Genre and crime rate over the years.
How would you achieve this task? Your answer should contain you strategy accompanied
with python code.

# Group by Genre and calculate the average crime rate for each genre
genre_crime_avg = merged_data.groupby('Genre')['IMDb Rating'].mean().reset_index()

# Merge the genre data with the crime rate data

merged_genre_crime = pd.merge(genre_crime_avg, crime_rate, on='Genre', how='left')

# Scatter plot to visualize the relationship

plt.figure(figsize=(10, 6))
sns.scatterplot(x='IMDb Rating', y='Crime rate', data=merged_genre_crime)
plt.title('Relationship between Movie Genre (IMDb Rating) and Crime Rate in the UK (2000-2022)')
plt.xlabel('Average IMDb Rating by Genre')

14
Roll#:

plt.ylabel('Crime Rate')
plt.show()

# Correlation analysis
correlation_matrix = merged_genre_crime[['IMDb Rating', 'Crime rate']].corr()
print("Correlation Matrix:\n", correlation_matrix)

Rough Sheet:

15
Roll#:

—---------------------Good Luck!----------------------------

16