You are on page 1of 10

Education management

machine learning model


Cristina Nițu
Victor Vlădăreanu
Ana-Maria Travediu
Alexandra Ciocîrlan
Vlad – Alexandru Grosu
Contents

 Background
 Raw data
 Variable modelling
 Pre-processing
 Training
 Testing
 Results and conclusions
Background

 Aim to investigate factors of educational achievement


 Data collected from school “Vintilă Brătianu”
 Anonymous survey of 122 students
 Cross – class (IX-XII), cross – disciplinary data
 Students on economic profil
Raw data

 Survey
 Obtain raw data
Variable Modelling

 9 considered independent variables


 Average grade always as dependent variable
 Missed classes can be switched
 Dependent variable for own model
 Independent variable for overall grade
Pre-processing

 Removed outliers
 Possible future clustering app
 Split multiple variables (sports question)
 Conversion of numerical variables to proper form (age)
 Conversion of two-class labels to binary
 Conversion of multi-class labels
 Ordinal
 Dummy coding
𝑥𝑖 −𝜇
 Normalisation 𝑥𝑖′ = 𝜎
Training

 Train / Validation / Test set split 70-20-10


 Models
 Linear Regression
 Polynomial Regression
 Neural Networks
 Algorithms
 Least squares
 Linesearch (fminunc)
 Gradient Descent
 Particle Swarm Optimisation
Testing

 Train / Validation / Test set split


 Set split procedure (70% - 20% - 10%)
 Validation of metaparameters
 Test results
Results and conclusions

 Test results – explained


 Model implementation – prediction and/or classification of educational results
 Future developments – more complex models, heuristic algorithms
 Benefit of further data gathering – reduce bias, allow complex model
 Future implementations of classroom robotics - Nao / expert systems
Thank you for your attention

You might also like