Criterion of the predictive potential | Comments | Refs. |
---|---|---|
\({R}^{2}=1-\frac{\sum {({Y}_{obs}-{Y}_{prd})}^{2}}{\sum {({Y}_{obs}-{\overline{Y} }_{train})}^{2}}\) | Yobs is the observed endpoint for the training set, and Ypred is the predicted endpoint values for the training set of compounds \({\overline{Y}}_{train}\) is the mean observed endpoint of the training set | Â |
\({Q}^{2}=1-\frac{\sum {({Y}_{prd(train)}-{Y}_{obs(train)})}^{2}}{\sum {({Y}_{obs(train)}-{\overline{Y}}_{train})}^{2}}\) | Yobs(train) is the observed endpoint, and Ypred(train) is the predicted response of the training set compounds | Â |
\(CCC=\frac{2\sum ({X}_{i}-\overline{X })({Y}_{i}-\overline{Y })}{\sum_{i=1}^{n}{({X}_{i}-\overline{X })}^{2}+\sum_{i=1}^{n}{({X}_{i}-\overline{X })}^{2}+n(\overline{X }-\overline{Y })}\) | n is the number of compounds, and xi and yi denote the mean of observed and predicted values, respectively | [51] |
\({\mathrm{c}}_{{R}_{p}^{2}}=R\sqrt{\left({R}^{2}-{R}_{r}^{2}\right)}\) | \({R}^{2}\) is squared correlation coefficient of models and \({R}_{r}^{2}\) is squared mean correlation coefficient of randomized models | [52] |
\(\overline{{r }_{m}^{2}}=\frac{{r}_{m}^{2}+{r}_{m}^{{\prime}2}}{2}\) | r2 is the squared correlation coefficient value between observed and predicted endpoint values, and \({\mathrm{r}}_{0}^{2}\) and \({\mathrm{r{\prime}}}_{0}^{2}\) are the respective squared correlation coefficients when the regression line is passed through the origin by interchanging the axes For the acceptable prediction, the value of all \(\Delta {\mathrm{r}}_{\mathrm{m}}^{2}\) metrics should preferably be lower than 0.2 provided that the value of r2m is more than 0.5 | [53] |
\(\Delta {r}_{m}^{2}=\left|{r}_{m}^{2}-{r{\prime}}_{m}^{2}\right|\) | ||
\({r}_{m}^{2}={r}^{2}\times \left(1-\sqrt{{r}^{2}-{r}_{0}^{2}}\right)\) | ||
\({r{\prime}}_{m}^{2}={r}^{2}\times \left(1-\sqrt{{r}^{2}-{r{\prime}}_{0}^{2}}\right)\) | ||
\(MAE=\frac{1}{n}\times \sum \left|{Y}_{obs}-{Y}_{prd}\right|\) | Â | Â |