SQL: Pushing a Dataset "Backwards"? - sql

Question

I have the following dataset that contains information on different students who took a fitness test, the date they took the fitness test, their weight at the time of the fitness test, and whether or not they passed the fitness test :
name date_test_taken result_of_test weight_after_time_of_test
1 john 2013-01-01 pass 165
2 john 2016-01-01 fail 183
3 john 2017-01-01 fail 175
4 john 2020-01-01 pass 182
5 alex 2019-01-01 fail 220
6 alex 2020-01-01 fail 225
7 tim 2018-01-01 pass 176
In this example, the student participates in the fitness test, then is told if they passed or failed, and then the student records their weight. The students don't necessarily take the test every year.
I am interested in building a statistical/machine learning model that will predict whether the student will pass or fail the NEXT fitness test they take based on the CURRENT weight of the student AND the result of their last fitness test.
This means if you take the second row of this dataset - John weighed 183 lbs after his second test, but his last known weight was actually 165 lbs. Therefore, I would be interested in "shifting" the dataset backward for each student. I am interested in predicting if John would have passed his second fitness test when his last known weight was 165 lbs and not 183 lbs.
Thus, using SQL code, I would like to "shift" the data for each student backward to modify the dataset. This way, the teacher can predict who will fail the next fitness test based on the results of the current fitness test - and then help those students more throughout the year.
Can someone please show me how to do this?
Thanks!

Answer 1

Obviously, with the information provided, we won't be able to provide you with the model to predict results. However, we can help with providing the data you need for your modelling.
The key task here is to get an individual's previous results into the same row (for analysis) as their current results.
SQL (at least SQL Server, and multiple other products) have functions LEAD and LAG that allow you to 'look forward' (lead) or backward (lag) through the dataset according to a given sorting and partitioning mechanism that tells it how to identify the previous/next rows.
In this case, we want to partition by the individual (name), take their previous result (LAG function, 1 row back) ordered by the date they took the test.
The following SQL gets the previous results for an individual onto the same row as the current data (note - I'm assuming the data table is called #FT and the first column is called 'Auto_ID'):
SELECT [name],
[date_test_taken],
LAG([date_test_taken], 1) OVER (PARTITION BY [name] ORDER BY [date_test_taken], [Auto_Id]) AS [date_test_taken_Previous],
LAG([result_of_test], 1) OVER (PARTITION BY [name] ORDER BY [date_test_taken], [Auto_Id]) AS [result_of_test_Previous],
LAG([weight_after_time_of_test], 1) OVER (PARTITION BY [name] ORDER BY [date_test_taken], [Auto_Id]) AS [weight_after_time_of_test_Previous],
[result_of_test],
[weight_after_time_of_test]
FROM #FT
Note that if they don't have a previous record, the previous results are NULL.
Here are the results
|name|date_test_taken|date_test_taken_Previous|result_of_test_Previous|weight_after_time_of_test_Previous|result_of_test|weight_after_time_of_test|
|----|---------------|------------------------|-----------------------|----------------------------------|----------------------------------------|
|alex|2019-01-01 |NULL |NULL |NULL |fail |220.00 |
|alex|2020-01-01 |2019-01-01 |fail |220.00 |fail |225.00 |
|john|2013-01-01 |NULL |NULL |NULL |pass |165.00 |
|john|2016-01-01 |2013-01-01 |pass |165.00 |fail |183.00 |
|john|2017-01-01 |2016-01-01 |fail |183.00 |fail |175.00 |
|john|2020-01-01 |2017-01-01 |fail |175.00 |pass |182.00 |
|tim |2018-01-01 |NULL |NULL |NULL |pass |176.00 |
To see it in action, here is a dbfiddle with the data, query, and results.