Skip to content

Project_Description.md

Latest commit

 

History

History
68 lines (59 loc) · 5.71 KB

Project_Description.md

File metadata and controls

68 lines (59 loc) · 5.71 KB

Smart Sales Customer Intelligence System - Detailed Project Description

1. Project Overview

The Smart Sales Customer Intelligence System is a machine learning-based web application built using Streamlit. It is designed to help businesses understand customer behavior, predict churn risks, segment customers into distinct groups, and forecast future customer spending. The system acts as a comprehensive dashboard where sales and marketing teams can input customer data (either manually or by selecting from an existing dataset) and receive actionable intelligence.

2. Core Features

The application offers three main predictive capabilities based on trained machine learning models:

  1. Customer Segmentation (Clustering): Groups customers into clusters (Group 0, 1, or 2) based on their profiles and purchasing behavior, allowing for targeted marketing strategies.
  2. Churn Risk Prediction (Classification): Predicts whether a customer is at "High Risk" or "Low Risk" of leaving (churning), enabling proactive retention efforts.
  3. Future Value Forecasting (Regression): Predicts the exact dollar amount a customer is likely to spend in the following month, helping in revenue forecasting.

3. Technology Stack

  • Frontend & Web Framework: Streamlit (app.py) for the interactive user interface.
  • Data Manipulation: Pandas, NumPy.
  • Machine Learning: Scikit-Learn (sklearn) for model building, preprocessing, and evaluation.
  • Data Serialization: joblib for saving and loading trained models and preprocessors.
  • Visualization (during training): Matplotlib, Seaborn.

4. System Architecture & Project Structure

  • app.py: The main application file containing the Streamlit UI, user input forms, logic to load models, data preprocessing steps for inference, and the final display of predictions.
  • Dataset/:
    • customer_data.csv: The raw dataset containing existing customer profiles and historical data.
    • preprocessed_data.csv: The cleaned and transformed dataset used for training the models.
  • Models/: Contains the scripts used to train the machine learning models.
    • Data_Preprocessing.py: Handles data cleaning, feature engineering, and splitting data into training and testing sets.
    • Classification_Model.py: Trains the Logistic Regression model for churn prediction.
    • Regression_Model.py: Trains the Linear Regression model for predicting next month's spend.
    • Unsupervised_model.py: Trains the KMeans clustering model using the Elbow pattern to find optimal segments.
  • pkl/: Stores the serialized machine learning artifacts required by the frontend application.
    • scaler.pkl: StandardScaler to normalize numerical features.
    • pca.pkl: Principal Component Analysis model for dimensionality reduction before passing data to models.
    • gender_encoder.pkl: LabelEncoder for the 'Gender' categorical variable.
    • Classification_Model.pkl: The trained Logistic Regression model.
    • Regression_Model.pkl: The trained Linear Regression model.
    • kmeans_model.pkl: The trained KMeans clustering model.

5. Machine Learning Models - Deep Dive

Data Preprocessing (Data_Preprocessing.py & app.py)

Before feeding data into the models, the system processes 9 key features:

  • Demographics: Age, Gender, Location (Urban, Suburban, Rural). Categorical variables are encoded (Label Encoding for Gender, One-Hot Encoding for Location).
  • Behavioral & Financial Data: Tenure (Months), Avg Monthly Spend, Last Month Spend, Num Transactions, Days Since Last Purchase, Support Tickets.
  • All numerical features are standardized using StandardScaler.
  • Data undergoes Principal Component Analysis (PCA) to reduce dimensionality while capturing the most important variance in the dataset.

A. Churn Classification (Classification_Model.py)

  • Algorithm: Logistic Regression.
  • Optimization: Hyperparameters (C, penalty, solver, class_weight) were tuned using RandomizedSearchCV with 5-fold Cross-Validation to find the best performing setup.
  • Output: Binary label deciding if a user is "High Risk" (1) or "Low Risk" (0).

B. Spend Regression (Regression_Model.py)

  • Algorithm: Linear Regression.
  • Optimization: Hyperparameters (fit_intercept, positive) were tuned using RandomizedSearchCV to minimize errors like Mean Absolute Error (MAE) and Root Mean Squared Error (RMSE).
  • Output: A continuous numerical value representing the estimated next month's spend.

C. Customer Segmentation (Unsupervised_model.py)

  • Algorithm: KMeans Clustering.
  • Optimization: The optimal number of clusters (k=3) was determined visually using the Elbow Method (plotting WCSS against the number of clusters) and evaluated using the Silhouette Score.
  • Output: Assigns the user to one of 3 distinct clusters.

6. User Workflow in the App

  1. Selection: The user opens the dashboard and chooses between "Existing Customer" or "Manual Entry" from the sidebar.
  2. Input:
    • If Existing Customer is chosen, selecting a Customer ID auto-populates all input fields using customer_data.csv.
    • If Manual Entry is chosen, the user manually fills in statistics like Age, Spend, Location, etc.
  3. Inference: Upon clicking "Predict":
    • The backend transforms the inputs precisely as the training data was (encoding, scaling, PCA).
    • The transformed array is passed to the three separate .pkl models simultaneously.
  4. Results: The dashboard displays the predicted Customer Segment (e.g., "Group 0"), Churn Risk status dynamically colored (Red for High Risk, Green for Low), and the exact predicted Future Value in dollars.