This report presents a Multiple Linear Regression analysis conducted to evaluate the impact of different marketing channels on sales performance. The objective is to determine how investments in TV advertising, Radio advertising, and Social Media marketing influence sales outcomes and to identify the most effective marketing channels for decision-making.

The analysis includes:

• Exploratory Data Analysis (EDA)
• Data preprocessing
• Multicollinearity assessment
• Multiple Linear Regression modeling
• Model diagnostics and validation
• Business interpretation and recommendations

Marketing organizations invest heavily in multiple advertising channels to drive customer engagement and increase sales. Understanding the effectiveness of each channel enables organizations to allocate resources efficiently and maximize return on investment (ROI).

This project uses Multiple Linear Regression to model the relationship between sales and three marketing channels:

• TV Advertising
• Radio Advertising
• Social Media Advertising

The dependent variable is:

• Sales

The independent variables are:

• TV Advertising
• Radio Advertising
• Social Media Advertising

Organizations need evidence-based insights to determine which marketing channels contribute most significantly to sales growth.

The key questions addressed are:

1. Which marketing channels significantly affect sales?
2. How strong is the relationship between marketing spend and sales?
3. Can sales be accurately predicted using marketing investment data?
4. Are there multicollinearity issues among predictors?
5. Does the regression model satisfy statistical assumptions?

The dataset contains observations of marketing expenditures and corresponding sales results.

The analysis follows these major stages:

1. Data Loading
2. Exploratory Data Analysis
3. Data Preprocessing
4. Multicollinearity Assessment
5. Regression Model Development
6. Model Diagnostics
7. Business Interpretation

The dataset was first examined using:

• Data information (info())
• Summary statistics (describe())
• Missing value analysis

• Understand data structure
• Identify missing values
• Detect potential anomalies
• Examine variable distributions

Descriptive statistics were generated to assess:

• Mean
• Median
• Standard Deviation
• Minimum Values
• Maximum Values
• Quartiles

These statistics provide an overview of data distribution and variability.

A missing value analysis was conducted to determine data completeness.

• Ensure model reliability
• Prevent biased results
• Determine if imputation is necessary

A pairplot was generated to visualize relationships among variables.

The pairplot helps identify:

• Linear relationships
• Nonlinear patterns
• Outliers
• Variable distributions
• Potential correlations

• Positive relationships between advertising channels and sales
• Distribution characteristics of each predictor
• Potential clustering patterns

The TV variable is categorical and must be converted into numerical form before regression analysis.

This transformation creates a new variable:

TV_Numeric

Regression models require numerical predictors. Encoding allows the TV advertising category to be incorporated into the model.

Multicollinearity occurs when predictors are highly correlated with each other.

High multicollinearity can:

• Distort coefficient estimates
• Increase standard errors
• Reduce model interpretability

Two methods were used to assess multicollinearity.

A heatmap of predictor correlations was generated using:

• TV_Numeric
• Radio
• Social Media

Correlation values range from:

-1 to +1

Where:

• +1 = Perfect positive correlation
• 0 = No correlation
• -1 = Perfect negative correlation

Strong correlations may indicate multicollinearity concerns.

VIF was calculated for each predictor.

VIF measures how much variance of a regression coefficient is inflated because of correlation with other predictors.

Ensure predictors contribute unique information to the model.

The regression model is defined as:

[ Sales = \beta_0 + \beta_1(TV_Numeric) + \beta_2(Radio) + \beta_3(Social\ Media) + \epsilon ]

Where:

• β₀ = Intercept
• β₁ = TV coefficient
• β₂ = Radio coefficient
• β₃ = Social Media coefficient
• ε = Error term

The model was fitted using:

• Ordinary Least Squares (OLS)
• Statsmodels library

OLS estimates coefficients that minimize the sum of squared residuals.

The regression summary provides several important metrics.

Measures the proportion of variance in sales explained by predictors.

Higher values indicate better model fit.

Adjusted R² accounts for the number of predictors.

Advantages:

• Penalizes unnecessary variables
• More reliable for multiple regression

Tests whether the model is statistically significant overall.

Null Hypothesis (H₀):

All coefficients are equal to zero.

Alternative Hypothesis (H₁):

At least one predictor significantly influences sales.

P-values determine predictor significance.

If:

p-value < 0.05

Then the predictor is considered statistically significant.

The TV coefficient measures the expected change in sales when TV advertising moves from one category level to another while holding other variables constant.

A positive coefficient indicates increased TV investment is associated with higher sales.

The Radio coefficient represents the expected change in sales for a one-unit increase in radio advertising expenditure.

A positive coefficient suggests radio advertising contributes positively to revenue generation.

The Social Media coefficient measures the impact of social media expenditure on sales.

A significant positive coefficient indicates social media campaigns effectively drive sales.

Regression assumptions must be validated before trusting model results.

A Q-Q plot was used to assess residual normality.

Residual points should closely follow the reference line.

Normal residuals improve confidence in:

• Hypothesis testing
• Confidence intervals
• Statistical inference

Residuals versus fitted values were plotted.

Random scatter around zero.

• Funnel shapes
• Curvature
• Systematic patterns

These may suggest:

• Heteroscedasticity
• Model misspecification

The residual plot also helps verify linear relationships between predictors and sales.

No visible pattern should exist.

Based on regression outputs:

1. TV advertising influences sales performance.
2. Radio advertising contributes to sales growth.
3. Social media advertising impacts sales outcomes.
4. Predictor significance is determined using p-values.
5. Multicollinearity is assessed through VIF scores.
6. Model validity is confirmed through diagnostic plots.

Allocate more budget toward channels with:

• High coefficients
• Statistically significant p-values

Track performance metrics regularly to ensure investments generate measurable returns.

Avoid dependence on a single marketing channel.

A diversified marketing strategy reduces risk and broadens customer reach.

Deploy the regression model for:

• Sales forecasting
• Budget planning
• Marketing optimization

The analysis may be limited by:

• Dataset size
• Variable selection
• Potential omitted variables
• External market conditions
• Assumption violations

Future studies may include additional predictors such as:

• Customer demographics
• Seasonality
• Competitor activity
• Economic indicators

This Multiple Linear Regression analysis demonstrates how marketing investments across TV, Radio, and Social Media channels can be used to explain and predict sales performance.

The methodology included exploratory analysis, preprocessing, multicollinearity assessment, regression modeling, and diagnostic testing. The resulting insights support data-driven marketing decisions and help organizations optimize advertising budgets for improved sales outcomes.

The model serves as a practical framework for understanding the relationship between marketing expenditure and business performance while providing a foundation for future predictive analytics initiatives.