Skip to content

Did you know that there are more than 190,000 fast-food restaurants in the US? In this project, you will explore sales of burgers and salads at two fast-food restaurants. The sales in burgers and salads have been picking up and they have had to place urgent orders to have enough stock since their previous predictions were wrong. This has costed them a lot of money!

These restaurants are interested in improving the daily predictions of their sales. They believe that calendar events, holidays and discounts have a high impact on the demand and are keen to know if they need to make changes to their inventory policies. Better predictions will help them fulfill more demand and reduce costs.

You will be working with historical food sales data. The data has been split into training and test sets (historicsales_fastfooditems_train.csv and historicsales_fastfooditems_train.csv) that contain the following columns:

  • restaurant - the name of the restaurant (chr)
  • item_name - the food item (chr)
  • date - the date of the sale (chr)
  • baseprice_USD - the price of the item before discount in USD (int)
  • discount_percent - the percentage discount (int)
  • sales_quantity - the quantity of item sold on a given date (int)
  • is_weekend - 1 if the sale date is Saturday or Sunday, otherwise 0 (int)
  • is_friday - 1 if the sale date is Friday, otherwise 0 (int)
  • is_holiday - 1 if the sale date is a holiday (that is not a weekend), otherwise 0 (int)
# Run this cell to load the libraries and data
# Load required libraries: dplyr, gplot2
library(dplyr)
library(ggplot2)

# Load the data
train <- read.csv("historicsales_fastfooditems_train.csv")
test <- read.csv("historicsales_fastfooditems_test.csv")

# Review the first few rows of your data
head(train)

# Check the data types
str(train)

# Fill in missing values in the discount_percent column
train$discount_percent[is.na(train$discount_percent)] <- 0
test$discount_percent[is.na(test$discount_percent)] <- 0

# Convert the date column into date format
train$date <- as.Date(train$date, format = "%d-%b-%Y")
test$date <- as.Date(test$date, format = "%d-%b-%Y")

head(train)

# Find correlation between the discount and sales quantity on the full training data
correlation <- cor(train$discount_percent, train$sales_quantity)
correlation

# Find the same correlation for different subsets of restaurant and food item
r1_burger <- train %>%
  filter(restaurant == 'R1' & item_name == 'Burger')

r1_burger_cor <- cor(r1_burger$discount_percent, r1_burger$sales_quantity)
r1_burger_cor 

r1_salad <- train %>%
  filter(restaurant == 'R1' & item_name == 'Salad')

r1_salad_cor <- cor(r1_salad$discount_percent, r1_salad$sales_quantity)
r1_salad_cor

r2_burger <- train %>%
  filter(restaurant == 'R2' & item_name == 'Burger')

r2_burger_cor <- cor(r2_burger$discount_percent, r2_burger$sales_quantity)
r2_burger_cor

r2_salad <- train %>%
  filter(restaurant == 'R2' & item_name == 'Salad')

r2_salad_cor <- cor(r2_salad$discount_percent, r2_salad$sales_quantity)
r2_salad_cor

# Save the highest correlation pair
highest_cor <- c('R1', 'Salad')
highest_cor

# Optional: plot time series data for each restaurant-item pair 
ggplot(train, aes(x = date, y = sales_quantity, color = interaction(item_name, restaurant, sep = "-"))) +
  geom_line() +
  labs(title = "Time Series of Sales Quantity by Restaurant and Item", x = "Date", y = "Sales Quantity (Units)", color = "Item, Restaurant")

# Build a regression model using the full training data
lm_sales_global <- lm(sales_quantity ~ discount_percent, data = train)

# Get the summary of the model
broom::glance(lm_sales_global)

# Try out different subsets of the data or predictors to get a better adjusted r-squared score. Here's one example:
lm_r1_burger_multi <- lm(sales_quantity ~ discount_percent + is_weekend + is_friday + is_holiday, data = r1_burger)

broom::glance(lm_r1_burger_multi)

# Store your best Adjusted R-Squared score
adj_rsquared = summary(lm_r1_burger_multi)$adj.r.squared

# Subset the test data to match training set
r1_burger_test <- train %>%
  filter(restaurant == 'R1' & item_name == 'Burger')

# Generate sales predictions using the test set
r1_burger_test$predicted_sales<- predict(lm_r1_burger_multi, newdata = r1_burger_test)

# Capture residuals for the data
r1_burger_test$residuals <- r1_burger_test$sales_quantity - r1_burger_test$predicted_sales

# Calculate the RMSE for the residuals
residuals <- r1_burger_test$sales_quantity - r1_burger_test$predicted_sales
residuals

rmse <- sqrt(mean(residuals^2))
rmse