Model Deployment Basics

Web Application for ML Predictions

Learning Outcome

6

Build a structured ML application using multiple Python modules.

5

Understand how Streamlit connects user input with ML predictions.

4

Understand the project structure of an ML prediction application.

3

Understand how trained models are reused using model serialization.

2

Explain how machine learning models are integrated into web apps.

1

Understand what a web application for ML predictions is.

Imagine building a powerful prediction engine.

It analyzes advertising budgets and forecasts sales.

But the engine lies inside a Python script.

Only you know how to run it.

To everyone else, it might as well not exist.

Now imagine placing that engine inside a simple interface.

Enter values.

Press a button.

The engine becomes a tool people can actually use.

 A web application transforms a machine learning model into a usable system.

A model is trained using a dataset.

  • Save trained model
  • serialized with Pickle

Model Storage

Model Training

User Interface

  • load the model

  • prepare input data

  • generate predictions

  • collects user input
  • Display prediction with steamlit

Prediction Logic

ML Web Application Architecture

Organizing code into multiple files improves readability and maintainability.

Each file performs a specific role in the application.

model.py — Model Training Module

This script is responsible for:

Loading the dataset

Training the machine learning model

Saving the trained model

Importing Libraries

These libraries handle:

import pandas as pd
from sklearn.model_selection import train_test_split
from sklearn.linear_model import LinearRegression
import pickle

Data Processing

Model training

Model Serialization

Loading Dataset

df = pd.read_csv(data_path)

The dataset contains advertising spending for:

 

 

 

 

 

 

The target variable is Sales.

Preparing Features

X = df[["TV", "Radio", "Newspaper"]]
y = df["Sales"]

Features represent advertising spending.

The target variable represents total sales.

Training the Model

model = LinearRegression()
model.fit(X_train, y_train)

The linear regression model learns how advertising budgets influence sales.

Saving the Model

with open(model_path, "wb") as f:
    pickle.dump(model, f)

The trained model is saved as linear_model.pkl

This saved file will later be loaded by the prediction system.

prediction.py — Prediction Logic

This module contains functions responsible for generating predictions.

Separating prediction logic keeps the application organized

Loading the Model

def load_model(model_path):
    with open(model_path, "rb") as f:
        model = pickle.load(f)
    return model

This function loads the serialized model file.

Prediction Function

def predict_sales(tv, radio, newspaper, model_path):
    model = load_model(model_path)
    features = np.array([[tv, radio, newspaper]])
    prediction = model.predict(features)[0]
    return prediction

  • User inputs are converted into numerical features.

  • The trained model predicts expected sales.

app.py — Streamlit User Interface

This script builds the interactive web application.

import streamlit as st
import pandas as pd
import matplotlib.pyplot as plt

Application Title

st.title("Advertising Sales Prediction App")

This displays the title of the application.

Displaying Dataset

if st.checkbox("Show raw dataset"):
    st.dataframe(df.head())

Users can optionally view the dataset.

Model Performance

st.write("R² Score:", r2_score(y, y_pred))
st.write("MSE:", mean_squared_error(y, y_pred))

These metrics show how well the model fits the dataset.

Visualization

ax.scatter(y, y_pred)

This plot compares actual sales with predicted sales.

User Input : Users provide advertising budgets using sliders.

tv = st.slider("TV Advertising ($)", 0.0, 300.0, 100.0)
radio = st.slider("Radio Advertising ($)", 0.0, 50.0, 20.0)
newspaper = st.slider("Newspaper Advertising ($)", 0.0, 120.0, 30.0)

Prediction

prediction = predict_sales(
tv, radio, newspaper, "linear_model.pkl")

The model predicts sales based on the input values.

Displaying Result

st.success(f"Predicted Sales: {prediction:.2f} units")

The predicted value is displayed on the webpage.

Summary

4

Interactive user interface

3

Prediction logic

2

Model serialization using pickle

1

Trained machine learning model

Quiz

What is the purpose of a web application for ML predictions?

A. Train models

B. Allow users to interact with models through a browser

C. Clean datasets

D. Improve accuracy

What is the purpose of a web application for ML predictions?

A. Train models

B. Allow users to interact with models through a browser

C. Clean datasets

D. Improve accuracy

Quiz-Answer

Web Application for ML Predictions

By Content ITV

Web Application for ML Predictions

  • 13

More from Content ITV