Script for prediction purchases based on Machine Learning algorithms.

Program uses:

• Pandas;
• Numpy;
• Scikit-Learn.

Program separated on two parts: preprocessing and prediction. Preprocessing includes next steps:

1. Removing duplicates from dataset.
2. Features selection
3. NaN Values processing: a. all missing number values replaced by 0; b. all missing text values replaced by word ‘miss’;
4. Processing date: a. Column ‘created_at’ replaced by days between today’s date and date in dataset; b. Column ‘ccreate’ replaced by ‘1’ or ‘0’ values (‘1’ - if cell in table not empty and ‘0’ - if cell is empty).
5. Convert categorical variable into dummy/indicator variables.

Using ML algorithms:

• Decision Tree;
• Random Forest;
• Gradient Boosting;
• Ensemble of those classifiers. All of them are binary classifiers, which at the end of script saving result to file ‘result.txt’.

Installing:

git clone https://github.com/exotikh3/behaviour_analysis.git

Then (python 2.7 required):

pip install -r requirements.txt

After this, please, sure than you have necessary datasets (‘outsource_test1.csv’ and ‘outsource_training.csv’) and then run the script. Take a coffee break). Each time model rebuilds and result writes down to ‘result.txt’.

Accuracy is on level between 72% (Decision tree) up to 80% (Ensemble and Gradient Boosting).