Kalhara JA← All Works
KJA·2026 · Case study
Tourism Arrival Predictor
Traditional ML forecasting with SHAP explainability
The brief
An end-to-end machine learning project that forecasts monthly tourist arrivals to Sri Lanka using SLTDA public data, compares traditional regression models, and provides explainable predictions through a Streamlit web app.
Approach
Chronological train/test splitting and time-series cross-validation across three model families; Random Forest won on RMSE/MAE/MAPE/R², then SHAP surfaced exactly which features drove each prediction.
Outcome
Shipped as a beginner-friendly Streamlit app with downloadable forecasts and in-app metric explanations — ML made legible to a non-technical user.
Key features
Public Sri Lankan Dataset Pipeline
Compiled monthly tourist arrival data (2016-2025) from SLTDA public reports with source traceability.
Traditional ML Model Comparison
Compared Ridge Regression, SVR (RBF), and Random Forest using chronological splitting and time-series cross-validation.
Best Model Evaluation
Selected Random Forest as best model based on validation/test metrics (RMSE, MAE, MAPE, R2).
SHAP Explainability
Generated SHAP global importance and summary plots to explain what drives model predictions.
Interactive Forecasting App
Built a beginner-friendly Streamlit app with scenario-based forecasting, downloadable CSV output, and in-app metric explanations.
Gallery