Sales Prediction Tool

This project is a Sales Prediction Tool that enables users to analyze and predict customer sales. It provides a fully customizable data preprocessing pipeline and machine learning hyperparameter tuning via a Flask and Angular interface.

Features

Data Analysis & Insights

• Exploratory Data Analysis (EDA) notebooks are available:
  • notebooks/data_analysis.ipynb
  • notebooks/EDA.ipynb
• Key analysis questions addressed:
  • Ordered plot of total transactions per customer (most to least active)
  • Monthly transaction frequency for a given product ID (for 2018)
  • Top 5 highest-selling products over the last six months
  • Data seasonality effect

Configurable Data Processing Pipeline

The pipeline consists of three transformation levels:

1. Level 1: Load and clean data (remove duplicates, transform date format)
2. Level 2: Compute monthly transactions per customer & define the target variable
3. Level 3: Compute rolling features and lag features

• Fully configurable via config/processing_config.json

Configurable Training Pipeline with Backtesting

• Ensures model robustness by simulating past performance
• Time-based data splits for training, validation, and testing:

Split	Training Period	Validation Period	Testing Period
Split 1	2017-01 to 2017-12	2018-01 to 2018-03	2018-04 to 2018-06
Split 2	2017-01 to 2018-03	2018-04 to 2018-06	2018-07 to 2018-09
Split 3	2017-01 to 2018-06	2018-07 to 2018-09	2018-10 to 2018-12
Split 4	2017-01 to 2018-09	2018-10 to 2018-12	2019-01 to 2019-03
Split 5	2017-01 to 2018-12	2019-01 to 2019-03	2019-04 to 2019-06
Split 6	2017-01 to 2019-03	2019-04 to 2019-06	2019-07 to 2019-09

• Various Machine Learning algorithms & clustering methods are supported
• Fully configurable via config/training_config.json
• Features can be selected via config/features.json

Configurable Inference Pipeline

• Forecasts sales using trained models
• Allows configuration of:
  • Number of models used for aggregation
  • Aggregation function (e.g., weighted method prioritizes recent models)
• Configurable via config/inference_config.json

User Interface (Flask & Angular)

• Web interface to edit configurations and run training pipeline

---

Getting Started

Prerequisites

Make sure you have the following installed:

• Python
• Node.js (Optional, for UI)
• Angular CLI (Optional, for UI)

Installation

1. Clone the repository:

   git clone https://github.com/Lahdhirim/ML-sales-prediction-tool.git
   cd ML-sales-prediction-tool

2. Install dependencies:
   • Python packages:

     pip install -r requirements.txt

   • Angular UI (Optional):

     cd src/config_editor/angular_ui
     npm install

---

Running the Pipelines

Three execution modes are available:

Run Processing & Training Pipeline (Without UI)

python main.py train

Or simply:

python main.py

Run Processing & Training Pipeline with Angular UI

python main.py train_with_ui

Run Inference Pipeline

python main.py inference

---

Output Files

Generated outputs are stored in data/output:

• raw_predictions.csv → Raw sales predictions from backtesting
• KPIs.xlsx → Performance metrics from backtesting
• inference.csv → Predictions during production phase