The City of Tampere’s recently completed artificial intelligence pilot will help to predict pupils’ need for special support in the future. The objective is to use automatic analysis of teachers’ notes and pupil information to better allocate school support resources.
The City of Tampere and IT service company CGI utilised machine learning to create a prediction model based on the data in the basic education pupil information system. The prediction model studied both pupil information and teachers’ notes and aimed to identify those pupils who are likely to need special support for attending school.
Thanks to the prediction model, pupils’ need for support will no longer be identified based only on teachers’ assessments. Instead, an analysis will be carried out on the basis of 193 different factors. Using the model, all pupils can be assessed in the same way.
The objective is to use data analysis so we can allocate and target support for pupils more equally than before, comments Education and Learning Director Kristiina Järvelä.
In the best case scenario, support targeted correctly and at an early enough stage can have a positive impact on an individual’s entire life.
How was the prediction model used in the data analysis created?
The model was made using pupil information in the Helmi system from 2013 to 2018. The information was anonymised so that individual pupils could not be identified. In the first phase, the machine learning algorithm was taught to identify factors that could indicate that the pupil would be assigned special support later on. The factors used in the data analysis included pupils’ grades in various subjects, absences from lessons, lesson notes, decisions to assign special support and pupil background information.
The algorithm created a model that was first tested with data from pupils who had received a decision for special support and then with data from those for whom decision information was not yet available. Based on the first data analysis, important factors were the average grade, grades in mathematics and Finnish language and whether or not the pupil had changed schools.
Prediction model strengthened by teachers’ unified note entry practices
Like all prediction models, the model used in the data analytics pilot makes mistakes and learns. In the future, the accuracy of the analytics model’s predictions can be increased either by combining the Helmi system data with data from other systems or by having teachers enter data in the system more uniformly.