This work contains a set of different models, that predict the price of Boston houses for Kaggle completion.
Short description of the dataset, that was used (https://www.kaggle.com/c/boston-housing ):
Feature
Description
Type
Example
crim per capita crime rate by town.
Numeric
0.00632
zn proportion of residential land zoned for lots over 25,000 sq.ft.
Numeric
18
indus proportion of non-retail business acres per town.
Numeric
2.31
chas Charles River dummy variable
Boolean
0
nox nitrogen oxides concentration (parts per 10 million).
Numeric
538
rm average number of rooms per dwelling.
Numeric
6575
age proportion of owner-occupied units built prior to 1940.
Numeric
65.2
dis weighted mean of distances to five Boston employment centres.
Numeric
4.09
rad index of accessibility to radial highways.
Numeric
1
tax full-value property-tax rate per $10,000.
Numeric
296
ptratio pupil-teacher ratio by town.
Numeric
15.3
black 1000(Bk - 0.63)^2 where Bk is the proportion of blacks by town.
Numeric
396.9
lstat lower status of the population (percent).
Numeric
4.98
medv median value of owner-occupied homes in $1000s.
Numeric
24
Outcome binary-classes models metrics:
Model name
Accuracy
Precision
Recall
F-measure
Logistic regression
91 %
91 %
91 %
91 %
Bagged decision tree
90 %
90 %
90 %
90 %
KNN classifier with 9 NN
89 %
91 %
90 %
90 %
Random forest with 70 trees
90 %
89 %
90 %
89 %
Linear regression
82 %
91 %
88 %
89 %
KNN classifier with 3 NN
80 %
89 %
84 %
85 %
Decision tree
76 %
82 %
76 %
79 %
Outcome 10-classes models metrics:
Model name
Accuracy
Precision
Recall
F-measure
Random forest with 580 trees
58 %
72 %
58 %
62 %
Bagged decision tree
55 %
69 %
55 %
59 %
KNN classifier with 3 NN
50 %
71 %
48 %
56 %
Logistic regression
43 %
55 %
43 %
46 %
Linear regression
73 %
41 %
31 %
30 %
Decision tree
36 %
40 %
36 %
35 %
Outcome clustering models metrics: