Prospects and Challenges of Using Machine Learning for Academic Forecasting

Table 2

Machine learning-based forecasting vs. traditional forecasting technique.

Machine learning forecasting	Traditional forecasting
It gives more accurate predictions with minimal loss function [10, 13]	Forecast errors are more likely to occur [38]
The approach is more scientifically driven [26]	Suffers a lot from assumptions leading to subjective conclusions at times [7]
Very demanding in computation [39]	Less demanding computation
It is more prone to underfitting and overfitting issues [40]	Less prone to underfitting and overfitting issues
Focuses more on result or outcome, but silent relationships among variables.	Relationship between variables are often highlighted
Highly recommended in applications where the goal is to learn from datasets with a large number of characteristics [41]	Suitable in univariate applications often meant to assess and summarize data.
It can work with massive data	It works with limited or historical data