Predicting Television Prices in Ghana: A Machine Learning Approach Using Weekly Market Survey Data

This paper presents the first machine learning price prediction model for the Ghanaian television market, built on nine weeks of structured weekly survey data (Weeks 18–26, 2026) covering seven television brands across three screen sizes (43", 50", and 55"). Using an Ordinary Least Squares (OLS) linear regression model with one-hot encoded brand features, screen size, and week number as predictors, the model achieves a cross-validated R² of 0.897 and a Mean Absolute Error (MAE) of GHS 666 — representing a prediction accuracy of approximately 88% (MAPE: 12.2%). Key findings include: brand identity explains approximately 90% of all price variance in the market; OLED televisions command a premium of GHS 7,772 above the model baseline; budget brands (Nasco, TCL) sit GHS 2,674–2,691 below it; and screen size contributes up to GHS 10,883 in price variation across the 43"–55" range. The paper includes five data visualisations covering model fit, residual analysis, brand price coefficients, predicted vs actual prices for Week 26, and a four-week price forecast (Weeks 27–30). A companion Kaggle notebook implementing Random Forest and XGBoost models is also available at the dataset DOI below. This work is part of the Ghana Market Price Index (SGMPI), an ongoing weekly price monitoring initiative by SG Datalytics. Dataset DOI: doi.org article DOI: doi.org: sgdatalytics.org