3.3. Identification of decision boundary by machine learning
Figure 7 shows the distribution of raw date set and identification of the decision boundary respectively for which the above data sets are desirable. This method can be further used to predict whether the testing data set should be termed good or poor (Bera, Saha, & Bhattacharjee, 2020). Logistic regression with regularisation gave the best output in classifying our experimental results (Shi, Li, Ding, & Gao, 2020). A lot of trails were done to set the values of regularisation constant as 1 and degree of feature mapping as 12. A decision boundary curve was plotted which divided the X-Y plane into two parts, one being the good yield region and the other poor (Figure. 7B) . The data points represented as “X” were termed as good yield and “O” as poor yield. It is evident from the graph that as the moisture content increases the lipid yield reduces. There is an increase in lipid yield when residual moisture content is reduced, drum temperature is increased and drum speed is reduced. This classification helps us in determining the range of the above mentioned factors in which one can get high lipid yield. The plot has an accuracy of 97.037% and corroborates our observation that a moisture content of <10% (wb) would result lipid recovery > 90% of the bone-dried biomass (Figure 7B).