Building a Salary Predictor with Scikit-Learn and Node.js

I wanted a project that demonstrates end-to-end ML: data preprocessing, model training, evaluation, and deployment. All running live on a 2GB VPS with zero Python dependencies at runtime.

The Problem

Salary data is noisy. The same "Data Scientist" title can pay anywhere from $60K to $250K depending on experience, location, company, and negotiation. Can we build a model that captures the broad trends?

Dataset

I used a Kaggle salary dataset with 6,700 records containing age, gender, education level, job title, and years of experience.

Preprocessing

• Filtered to tech/data roles using keyword matching (data, ML, engineer, developer, etc.)
• Stripped seniority prefixes (Senior, Junior, Lead) from titles since seniority is better captured by years of experience
• Standardized education levels (merged "Bachelor's Degree" into "Bachelor's", "phD" into "PhD")
• Kept titles with 20+ entries to avoid sparse categories

This reduced 52 raw titles down to 16 clean ones.

Model Comparison

I trained three models and compared them:

Model	R²	MAE
Linear Regression	0.65	$23,449
Random Forest	0.96	$4,375
Gradient Boosting	0.96	$4,576

Gradient Boosting won by a small margin. The ensemble models massively outperform linear regression because salary relationships aren't linear. A PhD with 10 years earns disproportionately more than a linear model would predict.

Deployment Strategy

The challenge: deploy an ML model on a 2GB VPS running Node.js with no Python, no pickle files, and no heavy dependencies.

Solution: Pre-computed predictions exported as JSON.

During training, I iterate over every valid combination of (job title x education x years of experience) and store the prediction. The resulting JSON is ~80KB, essentially a lookup table.

At runtime, Node.js loads the JSON once, and predictions are a simple key lookup. Zero computation, instant response.

function predict(jobTitle, education, yearsExp) {
  const key = `${jobTitle}|${education}|${yearsExp}`;
  return model.predictions[key];
}

Limitations

This model doesn't account for:

• Geographic location
• Company size or industry
• Equity, bonus, or total compensation
• Negotiation skills

It captures broad trends, useful for ballpark estimates but not offer negotiations.

Try It

→ Use the Salary Predictor