What did you contribute to the forecasting project?
Explain your work and check the result on your resume.
“The details given in the resume have to be genuine” (opens the source)
The question
You are interviewing for a full-time analytics trainee role through campus placement. You and three classmates built a demand forecast. A teammate wrote the model. You prepared the validation data and measured errors. An AI resume draft says you built the model and cut forecast error 30%.
| Value | |
|---|---|
| Validation cases | 20 |
| Baseline absolute error sum (units) | 400 |
| Model absolute error sum (units) | 280 |
Both totals cover the same validation rows. Mean absolute error is the total absolute error divided by the number of rows.
“Give a resume claim you can explain to the analytics panel.”
Not saved by this site. Your browser may restore text when you navigate back.
A strong answer, step by step
Open each step in turn. This is one strong answer, not the only one.
Step 1: Name your contribution
What would you do here? Open the move you’d make.
Credit the team model and describe your validation work.
The strong move. The team result and your own contribution answer different questions.
Use the draft’s claim that you built the model.
Tempting, but it misses something. The packet gives model authorship to your teammate and validation to you.
See how a strong candidate says it
“I prepared the validation data and measured forecast error. My teammate wrote the model. I can explain how we chose the baseline and how I checked the rows.”
Why it works: The team result and your own contribution answer different questions.
Step 2: Explain the result
What would you do here? Open the move you’d make.
Explain 20 versus 14 units of mean absolute error.
The strong move. A relative error reduction needs its baseline and sample.
Call it a 30 percentage-point accuracy gain.
Tempting, but it misses something. The calculation compares errors in units and expresses a relative reduction. It gives no accuracy rate.
See how a strong candidate says it
“On the reported 20 rows, the baseline mean absolute error was 20 units and the model’s was 14. That is 30% lower error relative to the baseline. I would name the metric and the validation set.”
Why it works: A relative error reduction needs its baseline and sample.
Step 3: Handle the leakage
The panel shows that exactly 4 evaluation records were also in training. The other 16 were held out from training and tuning. The four overlapping rows contribute 80 baseline and 24 model absolute-error units.
What would you do here? Open the move you’d make.
Subtract both overlapping sums and divide by 16.
The strong move. Use the same held-out rows for both methods.
Keep 20 as the denominator after dropping the four rows.
Tempting, but it misses something. The corrected error sums cover 16 records, and both methods must use those same records.
See how a strong candidate says it
“I would remove the four overlapping records from both totals. The remaining 16 rows give baseline error of 320/16 = 20 units and model error of 256/16 = 16 units. The reduction is now 20%.”
Why it works: Use the same held-out rows for both methods.
Step 4: Revise the claim
What would you do here? Open the move you’d make.
Report the corrected cohort and metric.
The strong move. State what survived the check and how you would improve the test.
Keep 30% because it was in the submitted resume.
Tempting, but it misses something. The 30% included four overlapping records. The corrected comparison supports 20% on the remaining sample.
See how a strong candidate says it
“I evaluated our team’s forecast and found 20% lower mean absolute error on the 16 remaining unseen records. I would use a locked future-period holdout for the next evaluation.”
Why it works: State what survived the check and how you would improve the test.
The answer, in short
- Decision
- Use the 20% error reduction with the cohort and your contribution.
- Because
- The corrected comparison covers 16 records held out from training and tuning.
- What would change it
- A locked future-period evaluation would give a stronger check.
Common traps
- Taking credit for a teammate’s model.
- Keeping a contaminated validation result.
Check your answer
- I named what I did.
- I gave the error metric and baseline.
- I removed the same rows from both methods.
- I revised the claim after the challenge.
Last reviewed Nov 10, 2026