Multiclass Forecasting on Panel Data Using Autoregressive Multinomial Logit and C5.0 Decision Tree

Panel data is commonly used for the numerical response variables, while literature forecasting categorical variables on panel structure still challenging to find. Forecasting important because it helpful government policies. This study aimed forecast multiclass or structure. The proposed models were autoregressive multinomial logit and C5.0. strategy applied so that two could be was add effects fixed predictor such as location, time, strata, month of observations. effect assumed a treated dummy variable. category land conditions through Area Sampling Frame (ASF) survey conducted by BPS-Statistics Indonesia. evaluation both based classification performance. Classification performance obtained dividing dataset into 75% training modeling 25% test validation then repeated 200 times. results showed C5.0 accuracy 86.48%, 83.97%. A comparison testing time sequence. result worse than Autoregressive had an 77.43%, 77.77%.