Python foundations · Exploratory data analysis

Understanding the FoodHub order experience.

FoodHub brings restaurants and customers together through a delivery app. I used order history to understand demand, ratings, restaurant performance, and delivery times—and to identify where the business could investigate next.

1,898 orders178 restaurants14 cuisinesCSV-backed results
01 / Data review

Start with the data we have

I checked the dataset’s size, field types, and rating completeness before comparing demand or performance.

1 · ProfileCheck rows, columns, types, and missing markers.
2 · DistributionsExplore restaurants, cuisine, prices, ratings, and times.
3 · RelationshipsCompare demand and operations across groups.
4 · AnswerTranslate the evidence into business answers and tests.

Data preparation decision

Ratings marked “Not given” are stored as text, not as true null values.

There are no literal nulls in the CSV. I counted 736 “Not given” ratings and kept them missing in the analysis. I don’t replace them with an average because that would create ratings customers never submitted and could distort restaurant averages.

02 / Univariate analysis

Explore each variable on its own

Order counts show where demand sits. Histograms and boxplots show the spread, typical range, and unusual values in cost and timing.

Order frequency by cuisine

All 14 cuisines, sorted by number of orders

Order frequency by restaurant

Top 10 restaurants; there are 178 in total

Order cost distribution

Histogram · dollars per order
Tukey boxplot · Q1, median, Q3, whiskers, and outliers

Ratings as recorded

“Not given” is shown separately, not scored as four stars

Food preparation time

Histogram · minutes from restaurant confirmation to pickup

Delivery time

Histogram · minutes from pickup to drop-off

Time spread at a glance

Tukey boxplots · minutes
The middle box spans Q1 to Q3; the line marks the median. Whiskers show the most extreme values within 1.5 × IQR.
03 / Bivariate & multivariate analysis

Compare variables together

These views show how order mix, prices, ratings, and time measures vary across groups. Small cuisine samples need extra caution.

Cuisine demand by day

Heatmap of order counts · rows are cuisine; columns are day group
Weekend order counts are higher for every cuisine in this dataset. The heatmap shows counts, so it also reflects the larger overall weekend volume.

Order price by cuisine

Boxplot · six most ordered cuisines · dollars
The box shows the middle 50% of orders; the center line is the median. Points beyond the whiskers are outliers.

Preparation time by recorded rating

Boxplot · submitted 3-, 4-, and 5-star ratings only

Delivery time by day

Boxplot · weekday vs weekend · minutes
Median delivery time is lower on weekends in this sample. This comparison does not establish why.

Preparation and delivery window

Heatmap · rows are delivery minutes; columns are preparation minutes

Mean delivery time by cuisine

All cuisines · count is included to show sample size
French and Vietnamese have the highest means here, but each has fewer than 20 orders. Treat those estimates as directional.
04 / Business questions

Answers for the business

The requested figures, recalculated directly from the CSV and using the stated definitions.

05 / Conclusions & recommendations

What I would take forward

The conclusions keep the original direction while separating what the data shows from what should be tested.

Recommendations

Test weekday incentives.

Reduce or adjust weekday delivery fees in a controlled promotion to encourage orders on slower days. Compare order lift, average order value, and delivery time with a baseline before expanding.

Promote niche restaurants during the week.

Try featuring less-frequented restaurants on weekdays, while making the most popular cuisines easy to find during busy periods. Track whether the placement brings new orders rather than shifting existing ones.

Investigate restaurants with few orders.

There are many restaurants with 20 or fewer orders. Look into discoverability, menu availability, pricing, and customer feedback before deciding which support would help.

Improve rating capture.

Follow up after delivery with a simple rating prompt. Monitor the response rate and the share of low ratings so missing feedback is not mistaken for positive feedback.

06 / Supporting data

Explore the complete dataset

All 1,898 source rows are packaged in this file. Search or filter the table, then export the data.

Order IDCustomer IDRestaurantCuisineCostDayRatingPrep (min)Delivery (min)
How the data is handled

The supplied CSV has 1,898 rows and 9 columns. “Not given” is a missing-rating marker. It remains separate from scores for every rating count, average, and promotion check.

Commission estimate

Applied 25% to order costs above $20 and 15% to order costs above $5 through $20. This is modeled commission, not profit; it excludes other expenses and revenue sources.