Application of Machine Learning Methods for the Classification of Discounts Types

This work is dedicated to the problem of supervised binary classification, where we explore 6 machine learning methods (Logistic Regression, Decision Tree, LGBM, CatBoost, Perceptron) and their ability to be applied to the dataset, containing detailed information about approved discounts for clients of a beer company. By training the models on a correctly labeled data, we aim to obtain the best classifier for predicting discounts types. Results show, that all the considering models have obtained significant improvement over baseline classifier (model which always returns majority label), with Random Forest having the best performance. Latter best classifier is planned to be used for restoring missed information about discount types, which is crucial for sales forecasting. Also, it turned out, that it is possible to train a model with practically acceptable performance using only 6 out of 17 features contained in the preprocessed dataset. The last fact potentially allows to restore information about missed discount types even for those periods where full training data isn’t available.