Data Science · Completed case study
A small transaction dataset becomes a disciplined exercise in data quality, statistical reasoning, customer behaviour and honest interpretation.
01 / Purpose
The aim was not to decorate a dataset with charts. It was to build a traceable analysis in which every result follows from a stated decision.
The project examines 281 retail transaction records across one month. It validates the data, handles missing sales transparently, studies the distribution of transaction values, tracks daily activity and measures customer contribution.
Analytical principle Report what the data supports—and make the boundaries of that evidence visible.
02 / Questions
Is the dataset complete, internally consistent and suitable for analysis?
What do the centre, spread and shape of transaction values reveal?
How do daily sales and transaction volume change across the month?
Which observed customers contribute the most recorded sales?
How is daily sales performance associated with transaction volume?
Which conclusions are supported—and which would overstate the evidence?
03 / Method
Each stage produces evidence used by the next.
Check fields, types, identifiers and missingness.
Parse dates and preserve the original data.
Build transaction, daily and customer views.
Inspect patterns, spread and influential values.
Separate observation from inference.
04 / Data quality
Four transactions have no sales amount. They are excluded from revenue calculations rather than filled with the mean, because imputation would manufacture sales that were never observed.
Decision: preserve the raw file, analyse complete sales records, and report the exclusion explicitly.
05 / Explore the month
Selected day
22 SeptemberSelect or focus a bar to inspect that day. Switch the measure to compare sales value with transaction volume.
06 / Findings
Each finding is paired with the qualification needed to interpret it responsibly.
The mean transaction value is 56.13, while the median is 36.28. A smaller group of high-value purchases pulls the mean upward, so the median better represents a typical transaction.
Interpretation: the six observations were retained because no available field indicates that they are data-entry errors.
Recorded daily sales range from 119.70 on 9 September to 1,147.62 on 22 September. The average daily total is 518.25.
Interpretation: one month can describe variation, but it cannot establish seasonality or support dependable forecasting.
Customer 48 generated the highest recorded sales: 524.41 across four transactions. The top ten customers account for 26.3% of the month’s total.
Interpretation: this is repeat activity inside September, not a long-term retention rate.
Daily transaction count and total daily sales have a Pearson correlation of r = 0.694, a moderately strong positive association.
Interpretation: daily sales is the sum of transaction values, so transaction count is mechanically part of the outcome. This is not causal evidence.
07 / Statistical perspective
A 10,000-sample percentile bootstrap estimates plausible ranges without assuming that the strongly skewed transaction values are normally distributed.
95% bootstrap interval
50.04 — 62.4595% bootstrap interval
26.09 — 45.5308 / Data
The compact structure is enough for transaction, daily and customer-level exploration. It is not enough for product, profitability or geographic analysis.
| Field | Type | Meaning |
|---|---|---|
customer_id | Integer | Anonymised customer identifier |
transaction_date | Date | Date of the transaction |
transaction_id | Integer | Unique transaction identifier |
sales | Numeric | Sales amount; currency not specified |
09 / Boundaries
One month cannot establish seasonality, trends or forecast reliability.
No product, category, channel, location, cost or profit fields are available.
Activity inside September cannot establish acquisition, churn or long-term retention.
The original source, sampling method and currency were not documented in the earlier dataset.
10 / Reproduce the work
The repository contains the dataset, guided notebook, reusable Python analysis, generated figures, summary tables and machine-readable results.
$ git clone https://github.com/
DrHiraBenish/sales-data-analysis.git
$ pip install -r requirements.txt
$ python src/analyze_sales.py
✓ Reports regenerated successfully.