Example : SVR

Problem Statement

Predict salary based on years of experience using Support Vector Regression.

Dataset

Step 1: Import Libraries

import pandas as pd
from sklearn.svm import SVR
from sklearn.preprocessing import StandardScaler

Step 2: Create Dataset

data = {
    "Experience": [1, 2, 3, 4, 5, 6],
    "Salary": [15000, 20000, 28000, 40000, 60000, 80000]
}

df = pd.DataFrame(data)

print(df)

Step 3: Separate Input and Output

X = df[["Experience"]]
y = df[["Salary"]]

Here:

• X = Experience

• y = Salary

Step 4: Apply Feature Scaling

SVR is sensitive to scale, so we scale both input and output.

scaler_X = StandardScaler()
scaler_y = StandardScaler()

X_scaled = scaler_X.fit_transform(X)
y_scaled = scaler_y.fit_transform(y)

Here:

• StandardScaler() creates a scaling object
• fit_transform(X) calculates mean and standard deviation, then scales the input values
• X_scaled contains normalized experience values

Step 5: Train SVR Model

model = SVR(kernel="rbf")

model.fit(X_scaled, y_scaled.ravel())

Here:

• kernel="rbf" helps handle nonlinear patterns.

• .ravel() converts target values into a 1D array.

Step 6: Predict Salary for 7 Years Experience

new_experience = [[7]]

new_experience_scaled = scaler_X.transform(new_experience)

predicted_salary_scaled = model.predict(new_experience_scaled)

predicted_salary = scaler_y.inverse_transform(
    predicted_salary_scaled.reshape(-1, 1)
)

print("Predicted Salary:", predicted_salary[0][0])

In this step, we predict the salary for a person with 7 years of experience. First, new_experience = [[7]] creates the new input value in 2D format because Scikit-Learn expects input data as rows and columns. Next, scaler_X.transform() converts the new experience value into scaled form using the same scaling applied during training. The SVR model then predicts the salary using this scaled input, and the result is stored in predicted_salary_scaled. Since the predicted value is still in scaled form, inverse_transform() converts it back to the original salary scale so that the result becomes understandable. Finally, print(predicted_salary[0][0]) displays the actual predicted salary value

Complete Program

import pandas as pd
from sklearn.svm import SVR
from sklearn.preprocessing import StandardScaler

# Dataset
data = {
    "Experience": [1, 2, 3, 4, 5, 6],
    "Salary": [15000, 20000, 28000, 40000, 60000, 80000]
}

df = pd.DataFrame(data)

# Input and output
X = df[["Experience"]]
y = df[["Salary"]]

# Feature scaling
scaler_X = StandardScaler()
scaler_y = StandardScaler()

X_scaled = scaler_X.fit_transform(X)
y_scaled = scaler_y.fit_transform(y)

# SVR model
model = SVR(kernel="rbf")
model.fit(X_scaled, y_scaled.ravel())

# Prediction
new_experience = [[7]]

new_experience_scaled = scaler_X.transform(new_experience)
predicted_salary_scaled = model.predict(new_experience_scaled)

predicted_salary = scaler_y.inverse_transform(
    predicted_salary_scaled.reshape(-1, 1)
)

print("Predicted Salary:", predicted_salary[0][0])

Output

Predicted Salary: around 68000 to 75000

The exact value may vary depending on the SVR parameters.

SVR first scales the data because salary values are much larger than experience values. Then it uses the RBF kernel to learn the nonlinear relationship between experience and salary. After prediction, the scaled salary is converted back to the original salary range using inverse_transform().

Summary

SVR is useful when the relationship between input and output is nonlinear. In this example, salary does not increase in a perfectly straight line, so SVR can model the pattern better than a simple linear model.