Machine Learning Model to Predict ROAS

BACKGROUND

A start-up launched marketing campaigns across media mix channels to help increase brand awareness and Click Through Rate (CTR) consideration towards students' enrollment. However, based on the historic data, Cost spent by the company is higher than the Return On Ads Spend (ROAS).

GOAL

Build a machine learning model to predict ROAS to help the marketing team make inform decision on how to allocate marketing budget.

METHODOLOGY

1. Performed exploratory data analysis on the historic data to help prepare the data derive and identify patterns from the data
2. Created at least 4 data visualizations to help identify trends and relationships visually 3.. Feature engineering a. Created new features for Click Through Rate (CTR), Cost Per Clicks (CPC) & Return Of Ads Spend (ROAS) b. Converted categorical variable into numeric variable using one-hot encoding
3. Built a predictive model to predict Return Of Ads Spend (ROAS) using Random Forest Algorithm

KEY INSIGHTS

1. There are no missing or null values and no duplicates
2. The columns with the strongest correlation are; clicks and spend because the result is 0.99 which is the number closest to 1.0
3. People aged 30-34 has the most clicks. There is more likelihood to get more enrollments from ages 30-34 and 25-29
4. Targeted marketing can be done for this age group (20-24) to convert these numbers to enrollments.
5. Strong positive correlation between clicks and spend
6. There is a relationship between the variables impressions, clicks and spend
7. Clicks are totally dependent on spend ( the more the marketing budget or spend the higher the likelihood of users clicking on the ads) clicks are totally dependent on impression and spend,
8. Impression and clicks have a very strong relationship of 1.0 . So it means more spend, more impression and clicks
9. Also conversion is highly dependent on impression, spent, clicks which already have a strong relationship

RECOMMENDATION

1. To increase the conversion rate, we need to increase the marketing budget because there is a strong correlation between marketing spend and conversion. The higher the budget, the higher the conversion.

MODEL VALIDATION

A value of 0 indicates no error or perfect predictions.

Mean Absolute Error is 0.0307 in range of 1, around 0 is best, this was pretty close. The lower the Mean Squared Error and close to 0, the better, this model attained around 0.0307

R2 Score close to 1 is best, here this model attained 0.74. Thus the prediction is 74% accurate since r2 is 0.74.

Root Mean Squared Error - provides an indication of the goodness of fit of a set of predictions to the actual values. This is a value between 0 and 1 for no-fit and perfect fit respectively. This model atttained around 0.25. The value of root mean squared error is 0.25, which is less than 10% of the mean value of the percentages i.e. This means that our algorithm did a decent job.