RANDOM FOREST FOR YIELD PREDICTION IN WHEAT (Triticum aestivum L.)

Abstract

An experiment was conducted at the central research farm, Department of Genetics and Plant Breeding (25˚ 13' latitude and 88˚ 23' longitude, and about 37.5 m above sea level) of Hajee Mohammad Danesh Science and Technology University (HSTU), Dinajpur to study Random forest for yield prediction in wheat (Triticum aestivum L.) during the period from December 2018 to April 2019. The experiment included ten genotypes namely BARI Gom 25, BARI Gom 26, BARI Gom 27, BARI Gom 28, BARI Gom 29, BARI Gom 30, BARI Gom 31, BARI Gom 32 and two advanced line BAW 1203 and BAW 1194. Data were recorded in days to 50 % germination,days to 50 % heading, plants per meter square, days to maturity, spikelet per meter square, plant height, tillers per plant, effective tillers per plant, grains per spike, spikelet per spikes, 1000 grain weight, yield per plant and yield per plot. All the data were statistically analyzed using the R-language software program. The results of the experiment revealed that there were a genetic variability and significant variations between genotypes were observed for all the characters. In case of genotypes BARI Gom 26 requires comparatively less day and BARI Gom 31 requires more days for 50% heading than other cultivars. Among th-e cultivars BARI Gom 27 is more longer whereas, BARI Gom 32 is comparatively shorter. The maximum thousand grain weight was recorded in BAW 1194 whereas the minimum thousand grain weight was recorded from BARI Gom 26. Random forest method was highly capable of predicting crop yields. The distribution of the number of nodes for the tree and it is 82.5. The highest yield/plant was found in BAW 1194. The lowest yield/plant found in BARI Gom 32. The highest yield/plot was found in BARI Gom 30 and the lowest yield found in BARI Gom 25. The number of grains/spike, has the highest value this mean it had a maximum importance in terms of contributing accuracy. Compared to them, it can be seen that at the effective tillers/plant contribution was negative. So this variable was not that important for prediction.