(The Gist of Science Reporter) How the Machine Learns

[August-2020]

“The dataset in our work is like a matrix with each row represented by material, and each column represents a specific feature of the material,” explains Arnab Kabiraj, first author of the study. Their dataset contained 157 materials, each with 1500 properties. “We then employed a 3-part method of processing,” he elaborates:
Select Percentile – the dataset columns are reduced by selecting the best features and discarding the irrelevant ones.
Zero Counts – all zero-valued data generated after step 1 are eliminated at this stage. And finally, the data is subjected to a standard supervised Machine Learning technique.
Gradient Boosting Regressor – an ensemble-based model that uses a predictor – like a decision tree – to enhance performance.