Development of models and information technology for predicting TV viewing metrics using data science methods

Accurate TV-metric forecasts are essential for broadcasters and advertisers. This study builds a data-science framework that predicts core measures—Rating, GRP, TRP and the Affinity Index—using panel people-meter data. Four models are compared on daily and weekly series: ARIMA, exponential smoothing (ETS), Prophet and an LSTM network. Parameters are autotuned, with performance tested by MAPE, RMSE and MAE. ARIMA and ETS lead (daily MAPE ≈ 10–13 %), and weekly aggregation improves accuracy further. By forecasting target and base GRPs we derive reliable affinity estimates, helping planners pick the most efficient channels and programmes and thus optimise campaign reach and cost-effectiveness.