/SalesPrediction

Primary LanguagePython

Sales Prediction

Description

This project was created to predict purchases in the online store and it consist three main file Classifier.py, Trainer.py and model.pkl. The first one is Python script that prepare data sets t=in format that is good for machine learning algorithms. The second one used to train algoritm for predecting and generate model.pkl. The third one is pre-trained machine learning model that was trained on data sets with absolutely same structure.

Data Praparation

Here is all steps for data preparation:

  1. Feature selection
  • sauce - store name
  • age_10pct - probability of cunsomer's age being in this range
  • age_25pct - probability of cunsomer's age being in this range
  • age_33pct - probability of cunsomer's age being in this range
  • age_50pct - probability of cunsomer's age being in this range
  • age_67pct - probability of cunsomer's age being in this range
  • age_75pct - probability of cunsomer's age being in this range
  • age_90pct - probability of cunsomer's age being in this range
  • gender_male_prob - probability of consumer being male based on name
  • gender_female_prob - probability of consumer being female based on name
  • no_first_name_data - consumer name unknown
  • agi_grp1_prob - probability of consumer's adjusted gross income in this range
  • agi_grp2_prob - probability of consumer's adjusted gross income in this range
  • agi_grp3_prob - probability of consumer's adjusted gross income in this range
  • agi_grp5_prob - probability of consumer's adjusted gross income in this range
  • agi_grp6_prob - probability of consumer's adjusted gross income in this range
  • no_income_data - income data for this individual is not available
  • us_state - US state
  • us_region - US region
  • ngeo - no account location identifier from geographic data
  • correct_first_name - first name info corrected
  • created_at - account creation date
  • ccreate - purchase date

  1. Drop duplicates

  2. Fill NaN

  3. Convert data into days

  4. Convert text features into vector of integers

Machine Learning Algorithm

For this task was used Desicion Tree from Scikit-Learn that return 71,44% of accuracy and work well and Stochastic Gradient Descent that return 81,12% of accuracy but the results isn't correct, because all data was labeled with no purchases. All another algorithm give us lower results for accuracy.

Technology

  • Python
  • Pandas
  • Scikit-Learn

Usage

To install all requirements, generate model.pkl via Trainer.py and run Classifier.py type the following command

pip install -r requirements.txt

python Trainer.py -f train.py

python Classifier.py -f test.csv