MLP-MCDA

Framework for ranking prediction based on Multi-Layer Perceptron (MLP) regressor model and historical datasets evaluated by experts using Multi-Criteria Decision Analysis (MCDA) methods in Python.

The main_ann.py file includes:

• Application of machine learning models from scikit-learn Python library:

  • MLPRegressor
  • LinearRegression

• And other methods:

  • GridSearchCV
  • cross_val_score
  • r2_score
  • train_test_split

• This framework uses the TOPSIS method from pyrepo-mcda Python package. You can install it via the pip command:

pip install pyrepo-mcda

• And Gini coefficient-based weighting method from crispyn Python package. You can install it via the pip command:

pip install crispyn

• Preparation of training and test datasets with feature values.
• Generation of the target variable representing MCDA score.
• Splitting dataset to train and test.
• Selection of the best hyper-parameters for MLP regressor model using GridSearchCV.
• Training and testing MLP regressor model in prediction rankings.
• Comparing MLPRegressor model with LinearRegression model.
• Determining the correlation between rankings.
• Results visualizations using column, line, scatter, and heat map.