Advertising Data Analytics – Sales Prediction Using Machine Learning

A complete machine learning project built to analyze advertising spend across TV, Radio, and Newspaper channels, and predict Sales using multiple regression models.

---

Dataset Overview

The dataset contains 200 advertising records with the following features:

• TV
• Radio
• Newspaper
• Sales (Target)

Key Details

• No missing values in the dataset
• Two outliers detected in the Newspaper column using the IQR method and removed
• Final dataset used for modeling: 198 rows × 4 columns

---

Data Preparation

Steps performed:

• Loaded CSV and verified shape (200 × 4)
• Confirmed 0 missing values
• Verified no duplicate entries
• Visualized distributions using boxplots
• Identified and removed outliers (IQR)
• Split dataset into training/testing (70% / 30%)
• Applied StandardScaler for feature normalization
• Separated feature variables (TV, Radio, Newspaper) and target (Sales)

---

Analysis & Visualizations

1. Data Distribution

• Boxplots showed wide variance in TV spending
• Newspaper included mild outliers
• TV, Radio, and Sales showed no outliers

2. Outlier Detection

• Only the Newspaper column contained 2 meaningful outliers

3. Train–Test Split

• Training samples: 138
• Testing samples: 60

4. Feature Scaling

• Standardization significantly improved model stability and accuracy

---

Machine Learning Models

Models trained after cleaning, scaling, and splitting:

Linear Regression

• MAE: 1.18
• RMSE: 1.58
• Accuracy: 91.39%

Decision Tree Regression

• MAE: 1.04
• RMSE: 1.53
• Accuracy: 91.75%

Random Forest Regression

• MAE: 0.93
• RMSE: 1.15
• Accuracy: 93.40%

XGBoost Regression

• MAE: 0.86
• RMSE: 1.17
• Accuracy: 93.90%

Cross-Validation Scores

Model	Mean CV Accuracy
Linear Regression	0.897
Decision Tree	0.853
Random Forest	0.927
XGBoost	0.934 (best)

---

Final Model

• XGBoost Regressor selected as final model
• StandardScaler used for feature normalization
• Model exported as hr.pkl
• Scaler exported as schr.pkl

---

Deployment

The project includes two user interfaces:

• Tkinter Desktop Application
• Streamlit Web Application

---

Key Insights

• TV and Radio advertising have the strongest positive impact on Sales
• Newspaper spending provides weak predictive power, especially after outlier removal
• Random Forest and XGBoost achieved high accuracy (~94%)
• Feature scaling improved model consistency
• Higher TV advertising budgets consistently correlate with higher sales

---

Technologies Used

• Python
• Pandas
• NumPy
• Scikit-learn
• XGBoost
• Seaborn
• Matplotlib
• Tkinter
• Streamlit

---

How to Run the Project

1. Clone the Repository

bash git clone https://github.com/yourusername/AdvertisingSalesPrediction.git

2. Install Dependencies

bash

Copy code

pip install pandas numpy matplotlib seaborn scikit-learn xgboost streamlit

3. Run the Notebook

bash

Copy code

jupyter notebook Advertising_Sales_Prediction.ipynb

4. Launch the Streamlit App

bash

Copy code

streamlit run hrapp.py

---

Future Improvements

• Add SHAP explainability for model interpretation

• Build an interactive dashboard (Streamlit or PowerBI)

• Perform hyperparameter tuning for each model

• Add advanced feature importance visualizations

• Deploy as a complete web application

---

Contributing

Pull requests are welcome.

You may enhance models, add visualizations, or extend deployment workflows.

---

Support

If you found this project useful, consider starring the repository.