Table 2 Comparison of traditional statistics AI/ML approaches.
From: Advancing data science research education in Africa through datathon-driven innovations
Characteristic | Analytical approach | |
|---|---|---|
Traditional statistics | AI/ML | |
Common name | Statistics | Machine learning |
Basis | Population inference based on a sample | Predictive models based on pattern analysis |
Starting points | Starting values, distributional assumptions | Training data |
General assumptions | Data distribution | Representative training data |
Sample size methodology | Widely available, commonly reported | Unstandardized, limited reporting |
Missing data | Permits missing data | Usually requires non-missing data |
Statistical significance | p-values, confidence intervals | Reliability, validation |
Classification tables | Contingency tables | Confusion matrices |
Prediction | Predictive models | Classification |
Analytics | Simple to complex | Complex |
Inferential strengths | Hypothesis testing, parameter estimation, and understanding relationships among predictors | Predictive precision, flexibility to accommodate complex data distributions |
Inferential weaknesses | Assumptions are not always fully justified | Black box approaches make intermediate processes difficult to understand |
