1
0
mirror of https://github.com/andre-wojtowicz/uci-ml-to-r.git synced 2024-11-28 16:25:27 +01:00

Added features statistics and raw numbers of class imbalance

This commit is contained in:
Andrzej Wójtowicz 2016-04-16 13:57:54 +02:00
parent 681bbca36c
commit 4da5529cae
2 changed files with 96 additions and 12 deletions

View File

@ -3,7 +3,7 @@ Andrzej Wójtowicz
Document generation date: 2016-04-16 01:21:55.
Document generation date: 2016-04-16 13:55:00.
@ -62,7 +62,16 @@ Document generation date: 2016-04-16 01:21:55.
```
**Class imbalance**: 11% / 89%
**Predictors**:
|Class | Frequency|
|:--------------|---------:|
|factor | 6|
|integer | 3|
|numeric | 5|
|ordered factor | 3|
**Class imbalance**: 11% / 89% (4254 / 33973)
---
@ -121,7 +130,13 @@ https://archive.ics.uci.edu/ml/citation_policy.html
```
**Class imbalance**: 37% / 63%
**Predictors**:
|Class | Frequency|
|:-------|---------:|
|numeric | 30|
**Class imbalance**: 37% / 63% (212 / 357)
---
@ -158,7 +173,13 @@ O. L. Mangasarian and W. H. Wolberg: "Cancer diagnosis via linear programming",
```
**Class imbalance**: 35% / 65%
**Predictors**:
|Class | Frequency|
|:-------|---------:|
|integer | 9|
**Class imbalance**: 35% / 65% (239 / 444)
---
@ -214,7 +235,16 @@ Ayres de Campos et al. (2000) SisPorto 2.0 A Program for Automated Analysis of C
```
**Class imbalance**: 22% / 78%
**Predictors**:
|Class | Frequency|
|:--------------|---------:|
|factor | 9|
|integer | 17|
|numeric | 2|
|ordered factor | 1|
**Class imbalance**: 22% / 78% (471 / 1655)
---
@ -264,7 +294,14 @@ Yeh, I. C., & Lien, C. H. (2009). The comparisons of data mining techniques for
```
**Class imbalance**: 22% / 78%
**Predictors**:
|Class | Frequency|
|:-------|---------:|
|factor | 3|
|integer | 20|
**Class imbalance**: 22% / 78% (6636 / 23364)
---
@ -302,7 +339,15 @@ https://archive.ics.uci.edu/ml/citation_policy.html
```
**Class imbalance**: 29% / 71%
**Predictors**:
|Class | Frequency|
|:-------|---------:|
|factor | 1|
|integer | 4|
|numeric | 5|
**Class imbalance**: 29% / 71% (167 / 416)
---
@ -341,7 +386,13 @@ https://archive.ics.uci.edu/ml/citation_policy.html
```
**Class imbalance**: 35% / 65%
**Predictors**:
|Class | Frequency|
|:-------|---------:|
|numeric | 10|
**Class imbalance**: 35% / 65% (6688 / 12332)
---
@ -383,7 +434,14 @@ Sikora M., Wrobel L.: Application of rule induction algorithms for analysis of d
```
**Class imbalance**: 7% / 93%
**Predictors**:
|Class | Frequency|
|:-------|---------:|
|factor | 4|
|integer | 11|
**Class imbalance**: 7% / 93% (170 / 2414)
---
@ -470,7 +528,14 @@ https://archive.ics.uci.edu/ml/citation_policy.html
```
**Class imbalance**: 39% / 61%
**Predictors**:
|Class | Frequency|
|:-------|---------:|
|integer | 2|
|numeric | 55|
**Class imbalance**: 39% / 61% (1813 / 2788)
---
@ -511,7 +576,14 @@ P. Cortez, A. Cerdeira, F. Almeida, T. Matos and J. Reis. Modeling wine preferen
```
**Class imbalance**: 37% / 63%
**Predictors**:
|Class | Frequency|
|:-------|---------:|
|factor | 1|
|numeric | 11|
**Class imbalance**: 37% / 63% (2384 / 4113)
---

View File

@ -64,11 +64,23 @@ for (dir.name in dir(PATH_DATASETS))
cat(str(dataset))
cat("\n```\n\n")
cat("**Predictors**:\n\n")
df.pred = data.frame(table(sapply(dataset[, 1:(ncol(dataset)-1)],
function(f){paste(class(f), collapse=" ")})))
colnames(df.pred) = c("Class", "Frequency")
cat(knitr::kable(df.pred, format="markdown"), sep="\n")
cat("\n")
perc.classes = sort(round(100*as.numeric(
table(dataset[, ncol(dataset)]))/nrow(dataset), 0))
num.classes = sort(as.numeric(table(dataset[, ncol(dataset)])))
cat(paste("**Class imbalance**:",
paste0(perc.classes[1], "% / ",
perc.classes[2], "%\n\n")))
perc.classes[2], "% (",
num.classes[1], " / ",
num.classes[2], ")\n\n")))
cat("---\n\n")
}
```