Predict Your Profits: Optimizing Ad Campaigns with Data-Driven Insights

CONFIDENTIAL: The information in this document belongs to Boston Institute of Analytics LLC. Any unauthorized sharing of this
material is prohibited and subject to legal action under breach of IP and confidentiality clauses.

Prediction Product AD
Campaign Performance
Presented by :Gravan Dmello

Introduction
• An advertising campaign is a set of advertisements that work together to promote a
product or service. An ad campaign is designed around a specific and unique theme to
create brand awareness about the company’s product or service.
• An advertising campaign can be a series of different individual ads or the same ad across
mediums used to create awareness and interest in a product or service.
• This is achieved through different forms of media, including radio, television, print
advertising, direct mail, or the internet.
• The intended objective of this project is to develop a robust supervised machine learning
model designed to accurately forecasts key performance indicators (KPIs) for future
product ad campaign. By achieving this goal, we aim to provide them with valuable
insights that can inform strategic decision-making, optimize resource allocation, and
enhance overall marketing effectiveness.

Dataset Information
Here are the key details about the dataset used in this project:
Our has 731 entries and 11 columns. The columns include
 There are three columns with float data types and eight with integer data types.
 The dataset has 5 categorical variables namely limit_infor, campaign_type, campaign_level,
product_level, resource_amount whose values have been represented by single digit numbers. Here,
the target variable is "orders"
limit_infor limits or restrictions associated with the marketing campaign or product.(0,1)
campaign_type type of marketing campaign, such as email, social media, print advertising,etc (0,1,2,3,4,5,6)
campaign_level level or scale of the marketing campaign, for example, national, regional, or local.(0,1)
product_level level or category of the product being marketed, such as high-end, mid-range, or budget.(1,2,3)
resource_amount resources (e.g., budget, personnel, or materials) allocated for the marketing campaign.(1,2,3,4,5,6,7,8,9)
email_rate email delivery rate or open rate.
price selling price of the product.
discount_rate discounts or promotional offers associated with the product.
hour_resources the number of labor hours or human resources dedicated to the marketing campaign or product sales efforts.
campaign_fee fees or costs associated with running the marketing campaign.
orders number of orders or sales generated for the product during the marketing campaign.

Exploratory Data Analysis (EDA)
• EDA is used to provides a provides a better understanding of data set variables and the
relationships between them.
• The dataset had no duplicates and 2 missing values in "price" which was 0.27% of total
data. Hence, the rows with missing values were dropped.
• While observing the relationship of numeric values with "orders", the campaign_fee
had one outlier which was removed for a cleaner data.

Exploratory Data Analysis (EDA)
• Correlation coefficient revealed that 'discount_rate' (0.232), email_rate'
(0.627), 'hour_resouces' (0.663), and 'campaign_fee’ (0.762) have positive correlations
with 'orders'. 'price' (-0.102) has a weak negative correlation with 'orders'.
• To ensure consistent scales for numerical features, MinMax Scaler was
employed during preprocessing.

Visualizations
• The independent variables "campaign_fee", "hour_resources" and "email_rate" have linear relationship with target
variable "orders" .

Visualizations
• There is a non-linear relationship between
price and the number of orders'
• The 'discount_rate' have positive relationship
with 'orders'.

• In this step, we divided the dataset into two parts: X and y.
X contains all the independent variables, which are the
features used to make predictions.
Meanwhile, y represents the dependent variable or target
variable, which is the outcome we want to predict.
• The dataset was split into training and testing sets.
• An 80:20 ratio was used, with 80% of the data allocated to
training and 20% to testing, and the test size set to 0.2.
• A random state of 42 was specified to ensure the
reproducibility of results across different runs
Splitting the data into X and y
Train-Test Split

Model Selection
The Prediction Product AD Campaign Performance is a
regression problem. Hence following models were used:
• Linear Regression is best for simple, linear relationships and
offers high interpretability.
• Support vector machine is versatile for both linear and non-
linear relationships but can be computationally expensive.
• Random Forest is powerful for complex, non-linear
relationships and provides robust performance but is less
interpretable and more computationally intensive.

Predictions:
Linear Regression Support Vector Machine
RMSE on Train Score: 0.04030 RMSE on Train Score: 0.04123
RMSE on Test Score: 0.04055 RMSE on Test Score: 0.04047
Difference between RMSE on
train and test set
0.00025
Difference between
RMSE on train and test
set
0.00076
Observation: Linear regression have shown slightly better results than SVM

Predictions:
Random Forest Regression
Before Tuning After Tuning
RMSE on Train Score: 0.0211 RMSE on Train Score: 0.0312
RMSE on Test Score: 0.0445 RMSE on Test Score: 0.0455
Difference between
set
0.0233
Difference between
set
0.0233
Observation: There is no significant improvement in RMSE values after hypertuning the
values

Feature Importance (Random Forest
Regression)
• 'campaign_fee' has the highest importance , 'hour_resources' have moderate importance and price have
minimal importance compared to other two.
• The remaining features (email_rate, discount_rate, campaign_type, resource_amount, product_level,
campaign_level, limit_infor) have negligible importance in the model.

The Best Performing ML Model

Conclusions
• The analysis and predictions provide valuable insights that can significantly enhance the ad
campaign performance for the company.
• Campaign Fee and Hour Resources: Increasing campaign fees and allocating more
resources positively influence the number of orders, suggesting that investments in these areas
are likely to yield higher returns.
• Pricing Strategy: Higher prices tend to reduce the number of orders. Therefore, maintaining
competitive and minimal prices can attract more customers and boost sales.
• Discount Rates: While higher discount rates can slightly increase the number of orders, their
impact is minimal. This indicates that focusing primarily on pricing and resource allocation
may be more effective than relying heavily on discounts.
• Model Performance: Linear regression outperforms SVM and random forest regression due
to the linear relationship between features and the target variable. This finding underscores the
importance of using a simpler, well-suited model to avoid overfitting and ensure accurate
predictions.
By leveraging these insights, the company can strategically allocate resources, optimize pricing,
and fine-tune their ad campaigns to maximize effectiveness and improve the overall return on
investment (ROI).

Predict Your Profits: Optimizing Ad Campaigns with Data-Driven Insights

More Related Content

Predict Your Profits: Optimizing Ad Campaigns with Data-Driven Insights