Salary Benchmark Prediction System
An end-to-end machine learning system for analysing workforce data and predicting salary benchmarks using multiple regression and predictive models.
- Type
- Machine Learning · Regression · Data Analysis
- Role
- ML Engineer
- Contribution
- Data preparation · Analysis · Modelling · Evaluation
The problem
Setting pay fairly and competitively depends on knowing what a role is worth, but salary benchmarks are often assembled from inconsistent surveys, job adverts and instinct. The factors that drive pay interact in ways that simple averages hide. Organisations need a benchmark that reflects those factors together and can be applied to a specific role. This project built a data-driven way to estimate that benchmark from workforce data.
What I built
An end-to-end machine learning system that takes raw workforce data through to salary predictions. The data is first cleaned and made consistent, handling missing values, inconsistent categories and outliers that would otherwise distort a salary model.
Exploratory analysis then examines how pay relates to each available attribute, which informs the features the models receive. Several regression and predictive models are trained on the same prepared data and compared on held-out records using standard error metrics, so the choice of final model rests on evidence rather than preference. The selected model produces salary benchmark predictions for new profiles, with the preceding analysis documenting why it behaves as it does.
Approach
The full workflow, not just the prediction
Key decisions
Clean before modelling
Missing values, inconsistent categories and outliers are resolved first, because a salary model trained on untidy data learns the untidiness.
Let the analysis choose the features
Exploratory analysis decides which attributes and derived variables reach the models, rather than feeding every column in.
Compare models on the same split
Every model is trained and scored on identical data and metrics, so differences in the results reflect the models, not the data they saw.
Outcome
A complete, reproducible workflow from raw workforce data to salary benchmark predictions, with each stage, cleaning, analysis, feature engineering and model comparison, documented so the final prediction can be explained as well as produced.
- Capabilities
Working on something similar?
Start a conversation