Hedaro

Search

SearchSearch

Data Scientist - Restaurant Revival Challenge

Sep 15, 2024, 8 min read

Note

Description:

You’re the Data Scientist hired by a popular restaurant chain to analyze their sales data and help them optimize their menu and marketing strategies. The chain has 500 locations across the US, each offering a unique blend of local and national menu items. Your task is to clean, preprocess, and analyze the sales data to identify trends, patterns, and insights that can drive business growth.

Tasks:

  • Menu Engineering: Create a new feature that calculates the average sales per menu item across all locations. Use this feature to identify the top 10 best-selling menu items and the bottom 10 worst-selling menu items.
  • Location Analysis: Merge the sales data with a separate location data frame to create a single data frame with location-specific sales data. Then, use pivoting to calculate the total sales for each location and the top 3 best-selling menu items at each location.
  • Seasonal Trends: Use data transformation to convert the sales data into a weekly format (i.e., sales per week of the year). Then, use reshaping to calculate the average sales per week for each menu item and identify the peak sales periods for each item.
# import libraries
import pandas as pd
import numpy as np
import sys
 
print('Python version ' + sys.version)
print('Pandas version ' + pd.__version__)
print('Numpy version ' + np.__version__)
Python version 3.11.7 | packaged by Anaconda, Inc. | (main, Dec 15 2023, 18:05:47) [MSC v.1916 64 bit (AMD64)]
Pandas version 2.2.1
Numpy version 1.26.4

The Data

The dataset consists of 100,000 sales records from a restaurant chain with 500 locations across the US, detailing sales amounts, dates, menu items, and location information. It provides a comprehensive snapshot of sales activity across various menu items, locations, and time periods.

Columns:

  • sale_id: Unique sale identifier
  • menu_item_id: Unique menu item identifier
  • location_id: Unique location identifier
  • sale_date: Date of sale
  • sales_amount: Total sales amount
  • item_name: Menu item name
  • category: Menu item category (Appetizer, Entree, Dessert)
  • city: Location city
  • state: Location state
# set the seed
np.random.seed(0)
 
# generate menu item data
menu_items = pd.DataFrame({
    'menu_item_id': range(1, 101),
    'item_name': [f'Menu Item {i}' for i in range(1, 101)],
    'category': np.random.choice(['Appetizer', 'Entree', 'Dessert'], 100)
})
 
# generate location data
locations = pd.DataFrame({
    'location_id': range(1, 501),
    'city': [f'City {i}' for i in range(1, 501)],
    'state': np.random.choice(['CA', 'NY', 'TX', 'FL'], 500)
})
 
# generate sales data
sales = pd.DataFrame({
    'sale_id': range(1, 100001),
    'menu_item_id': np.random.randint(1, 101, 100000),
    'location_id': np.random.randint(1, 501, 100000),
    'sale_date': pd.date_range('2022-01-01', '2022-12-31', periods=100000),
    'sales_amount': np.random.uniform(10, 100, 100000)
})

Menu Engineering:

Create a new feature that calculates the average sales per menu item across all locations. Use this feature to identify the top 10 best-selling menu items and the bottom 10 worst-selling menu items.

We need the sales and menu_items dataframes for this task. Let us look at the data types for each dataframe.

menu_items.info()
<class 'pandas.core.frame.DataFrame'>
RangeIndex: 100 entries, 0 to 99
Data columns (total 3 columns):
 #   Column        Non-Null Count  Dtype 
---  ------        --------------  ----- 
 0   menu_item_id  100 non-null    int64 
 1   item_name     100 non-null    object
 2   category      100 non-null    object
dtypes: int64(1), object(2)
memory usage: 2.5+ KB

sales.info()
<class 'pandas.core.frame.DataFrame'>
RangeIndex: 100000 entries, 0 to 99999
Data columns (total 5 columns):
 #   Column        Non-Null Count   Dtype         
---  ------        --------------   -----         
 0   sale_id       100000 non-null  int64         
 1   menu_item_id  100000 non-null  int32         
 2   location_id   100000 non-null  int32         
 3   sale_date     100000 non-null  datetime64[ns]
 4   sales_amount  100000 non-null  float64       
dtypes: datetime64[ns](1), float64(1), int32(2), int64(1)
memory usage: 3.1 MB

Here is my strategy to complete the task:

  • Merge the sales and the menu_items dataframes
  • Create a group object and determine the average sales per menu
  • Order the data to select the top/bottom 10
# we are doing an inner merge where both dataframes must have matching menu_item_ids
df = sales.merge(menu_items, on='menu_item_id')
df.head()
sale_idmenu_item_idlocation_idsale_datesales_amountitem_namecategory
0127172022-01-01 00:00:00.00000000065.635255Menu Item 27Appetizer
12974962022-01-01 00:05:14.49914499150.715654Menu Item 97Dessert
23521332022-01-01 00:10:28.99828998239.340689Menu Item 52Entree
34741302022-01-01 00:15:43.49743497422.613624Menu Item 74Appetizer
45545002022-01-01 00:20:57.99657996528.305376Menu Item 54Entree

Notice the two columns in the results above (item_name and category); these came from the menu_items dataframe.

Now that the merge is complete, the code below will get us the average sales amount per menu item.

# create group object
group = df.groupby('item_name')
 
# get the average sales_amount
avg_sales = group['sales_amount'].mean()
avg_sales.head()
item_name
Menu Item 1      56.313134
Menu Item 10     55.226327
Menu Item 100    54.597751
Menu Item 11     53.615531
Menu Item 12     55.214489
Name: sales_amount, dtype: float64

Order the data to get the top/bottom 10 rows.

# top 10
avg_sales.sort_values(ascending=False).head(10)
item_name
Menu Item 99    57.838223
Menu Item 57    57.167071
Menu Item 8     56.666762
Menu Item 75    56.603161
Menu Item 24    56.436106
Menu Item 61    56.349409
Menu Item 1     56.313134
Menu Item 86    56.188104
Menu Item 84    56.018658
Menu Item 74    56.018637
Name: sales_amount, dtype: float64

# bottom 10
avg_sales.sort_values().head(10)
item_name
Menu Item 32    53.261398
Menu Item 88    53.599913
Menu Item 23    53.607014
Menu Item 62    53.612730
Menu Item 11    53.615531
Menu Item 33    53.620281
Menu Item 4     53.830895
Menu Item 81    53.931528
Menu Item 66    53.957063
Menu Item 15    54.032607
Name: sales_amount, dtype: float64

Location Analysis:

Merge the sales data with a separate location data frame to create a single data frame with location-specific sales data. Then, use pivoting to calculate the total sales for each location and the top 3 best-selling menu items at each location.

Our last task merged sales data with menu items data and placed it in df. We can reuse the dataframe for this task.

# add in the location data
location = df.merge(locations, on='location_id')
location.head()
sale_idmenu_item_idlocation_idsale_datesales_amountitem_namecategorycitystate
0127172022-01-01 00:00:00.00000000065.635255Menu Item 27AppetizerCity 17NY
12974962022-01-01 00:05:14.49914499150.715654Menu Item 97DessertCity 496CA
23521332022-01-01 00:10:28.99828998239.340689Menu Item 52EntreeCity 133CA
34741302022-01-01 00:15:43.49743497422.613624Menu Item 74AppetizerCity 130FL
45545002022-01-01 00:20:57.99657996528.305376Menu Item 54EntreeCity 500CA
# create group object
group = location.groupby(['city','state','item_name'])
 
# calculate the total sales for each group
total_sales = group['sales_amount'].sum()
total_sales
city     state  item_name    
City 1   TX     Menu Item 1      370.460969
                Menu Item 10     271.469363
                Menu Item 100    134.266236
                Menu Item 11     243.996257
                Menu Item 12     168.741860
                                    ...    
City 99  NY     Menu Item 94     100.391216
                Menu Item 96     141.602043
                Menu Item 97      59.668432
                Menu Item 98     295.399819
                Menu Item 99      68.530342
Name: sales_amount, Length: 43135, dtype: float64

Method 1:

Here is how I was able to get the top 3 menu items per location using a custom function.

Notes:

  • Performing an aggregate function on a group is pretty simple. Performing an additional calculation afterwards, this is a bit harder.
  • Since I reset the index, it kind of got in the way, so I added code specifically to remove this column (I used droplevel() to accomplish this)
def get_top_three(grp):
    ''' sort grp descending and return the top 3 rows '''
    sorted = grp.sort_values('sales_amount', ascending=False)
    return sorted.head(3)
 
# reset the index before performing another groupby operation
# group by city and state
# apply the get_top_three function on the group
results = total_sales.reset_index().groupby(['city','state']).apply(get_top_three, include_groups=False)
 
# get rid of number column in the index
results.index = results.index.droplevel(2)
results.head()
item_namesales_amount
citystate
City 1TXMenu Item 86419.518108
TXMenu Item 1370.460969
TXMenu Item 30342.247197
City 10TXMenu Item 63300.661903
TXMenu Item 40298.190699

Method 2:

Here is how I was able to get the top 3 menu items per location using a lambda function.

See the number column in the index? It’s the 3rd one after the state column. This column was created after the reset_index() operation. Method #1 shows you code on how to get rid of it.

I prefer Method #1 as that method uses cleaner code and it will be easier to maintain.

results = total_sales.reset_index().groupby(['city','state']).apply(lambda x: x.sort_values('sales_amount', ascending=False).head(3), include_groups=False)
results.head()
item_namesales_amount
citystate
City 1TX73Menu Item 86419.518108
0Menu Item 1370.460969
20Menu Item 30342.247197
City 10TX133Menu Item 63300.661903
113Menu Item 40298.190699

Seasonal Trends:

Use data transformation to convert the sales data into a weekly format (i.e., sales per week of the year). Then, use reshaping to calculate the average sales per week for each menu item and identify the peak sales periods for each item.

I will be using the Pandas resample method to convert the data to weekly data. Note that this method expects out data to be a time series.

Here is what I look for to determine if I am working with time series data:

  • No missing values
  • Dates are ordered sequentially
  • Dates occur at consistent intervals

Note that our synthetic data does not meet the typical criteria for time series data, as it does not contain consistent intervals. Keep this in mind when working through this tutorial.

# let us order the location data
location = location.sort_values(by='sale_date')
 
# Note: We don't have any missing values. Unfortunately our dates do not occur at consistent intervals
 
# resample our data to weekly and calculate the average sales_amount
trends = location.groupby('item_name').apply(lambda x: x.resample('W', on='sale_date')['sales_amount'].mean(), include_groups=False)
trends.head()
item_name    sale_date 
Menu Item 1  2022-01-02    58.064839
             2022-01-09    52.142763
             2022-01-16    49.412837
             2022-01-23    58.902413
             2022-01-30    59.432552
Name: sales_amount, dtype: float64

Using the same strategy for the Location Analysis, we can get the peak sales date and sales amount per menu item.

def get_peak_sales(grp):
    sorted = grp.sort_values('sales_amount', ascending=False)
    return sorted.head(1)
 
results = trends.reset_index().groupby('item_name').apply(get_peak_sales, include_groups=False)
 
# get rid of number column in the index
results.index = results.index.droplevel(1)
results.head()
sale_datesales_amount
item_name
Menu Item 12022-07-2467.081441
Menu Item 102022-01-0271.869690
Menu Item 1002022-01-0287.621319
Menu Item 112022-09-1166.620358
Menu Item 122023-01-0172.514490

Summary

You’ve completed a tutorial on sales data analysis for a 500-location US restaurant chain. You explored menu engineering, location analysis, and seasonal trends to optimize menu and marketing strategies, gaining practical insights into data-driven business decisions.

Key Takeaways:

  • Merging datasets using pd.merge()
  • Grouping and aggregating data using df.groupby()
  • Identifying top and bottom performers using sort_values() and head()
  • Using custom functions and lambda functions for complex aggregations
  • Resampling time series data using pd.resample() to convert to weekly format

Graph View

Backlinks

  • No backlinks found