Rice production in Japan is facing problems of yield and quality instability owing to recent climate changes and a decline in rice prices, and possible competition with foreign inexpensive rice. Thus, it is becoming more important to stably achieve high yield and quality, while reducing production costs. Various data, including crop growth, farmer’s management styles, yield and quality, has recently become accessible in actual fields using advanced information and communication technologies. Those data can be effectively used to aid farmer’s decision-making on their management. In this study, we built predictive models of brown rice yield (yield) using 85 data sets collected from 2010 to 2015 at 21 paddy fields in Itoshima city, Fukuoka prefecture, Japan. In the paddy fields, rice was cultivated under various environmental conditions and management styles. Support vector machine was applied to build the models to predict three yield classes (low, middle, high) that ranged from 2.98 to 6.17 t ha^-1. The models were built using all the combinations of four explanatory variables: number of spikelets, inorganic nitrogen (N) supply from panicle initiation to mid-ripening dates, and average values of sunshine hours and air temperature from heading to mid-ripening dates. A yield predictive model had the highest classification accuracy of 79% when the model selected two variables: the number of spikelets and inorganic N supply. The model identified high-yield conditions ranging from 27,216 to 30,146 m^-2 for number of spikelets and from 58.1 to 72.1 kg ha^-1 for inorganic N supply. The results indicated that shortage of inorganic N supply was one of the major causes to lower yield. The results also suggested that applying second topdressing was necessary to achieve high yield in the target region.