# Investigating the Predictive Performance of Process Data and Result Data in Complex Problem Solving Using the Conditional Gradient Boosting Algorithm

## TL;DR

This study uses a machine learning algorithm to compare how well process and result data predict complex problem-solving skills in students.

## Contribution

The study introduces the use of conditional gradient boosting to evaluate the predictive power of process and result data in problem-solving assessments.

## Key findings

- Process data showed moderate prediction performance, while result data showed moderate-to-good performance.
- Combining process and result data improved prediction performance significantly.
- Mathematical literacy and the VOTAT strategy score were the most influential predictors.

## Abstract

This study aims to examine the predictive performance of process data and result data in complex problem-solving skills using the conditional gradient boosting algorithm. For this purpose, data from 915 participants of the 2012 cycle of the Programme for International Student Assessment (PISA) were utilized. Process data were obtained from the log file of the first question in the climate control unit task included in the problem-solving assessment of PISA 2012. Various cognitive and affective attributes from the same assessment were used as the result data. According to the results, (1) process data demonstrated a moderate, result data demonstrated a moderate-to-good, and process + result data demonstrated a good prediction performance. (2) The most effective variables were the VOTAT (vary-one-thing-at-a-time) strategy score and total time in process data; the mathematical literacy and reading literacy scores in result data; and the mathematical literacy and VOTAT strategy score in process + result data. The dominance of the mathematical literacy has been noteworthy.