Danone Hackathon - Neural Network

The task is to predict future sales during the unstable Covid-19 period.

Steps:

• One-Hot Encoding for categorical variables
• Distribution analysis and removing outliers
• Splitting to train, validation and test
• Standard Scaler for normalizing data
• Converting to tensors PyTorch
• We build and train a model (MLP), define an optimizer and a loss function
• Analyze training indicators and save the model

Kaggle Competition - Logistic regression & XGBoost

The task is to forecast the number of goods which will be sold in each shop next month.

Steps:

• Data preprocessing and adding features
• Adding lag features (since time-series data)
• Splitting to train, validation and test
• Build and test models:
  • Logistic regression
  • XGBoost with cross-validation to select the best hyper-parameters
• Compare the accuracy of the models:
  • RMSE, Logistic regression = 1.25
  • RMSE, XGBoost = 1.23