Machine Learning

# Random Forest for Car Quality[Case Study]

Find your way out of the Data Forest with Random Forest

## Introduction :

In this blog we will discuss one of the most widely used Ensembling Machine Learning Algorithm called Random Forest. The goal of the blogpost is to get the beginners started with fundamental concepts of a Random Forest and quickly help them to build their first Random Forest model.

Motive to create this tutorial is to get you started using the random forest model and some techniques to improve model accuracy. In this article, I’ve explained the working of random forest and bagging.

Random forest is a tree-based algorithm which involves building several trees (decision trees), then combining their output to improve generalization ability of the model. The method of combining trees is known as an ensemble method.

Ensemble methods are supervised learning models which combine the predictions of multiple smaller models to improve predictive power and generalization.

Say, you want to buy a car. But you are uncertain of its quality. You ask 20 people who have previously . 12 of them said ” the car is excellent.” Since the majority is in favor, you decide to go for it. This is how we use ensemble methods in machine learning too.

The smaller models that combine to make the ensemble model are referred to as base models. Ensemble methods often result in considerably higher performance than any of the individual base models could achieve.

## Two popular families of ensemble methods

BAGGING

Several estimators are built independently on subsets of the data and their predictions are averaged. Typically the combined estimator is usually better than any of the single base estimator.

Bagging can reduce variance with little to no effect on bias.

ex: Random Forests

BOOSTING

Base estimators are built sequentially. Each subsequent estimator focuses on the weaknesses of the previous estimators. In essence several weak models “team up” to produce a powerful ensemble model. (We will discuss these later this week.)

Boosting can reduce bias without incurring higher variance.

## Conditions for ensembles to outperform base models

For an ensemble method to perform better than a base classifier, it must meet these two criteria:

1. Accuracy: the combination of base classifiers must outperform random guessing.
2. Diversity: base models must not be identical in classification/regression estimates.

## Bagging

The ensemble method we will be using today is called bagging, which is short for bootstrap aggregating.

Bagging builds multiple base models with resampled training data with replacement. We train k base classifiers on n different samples of training data. Using random subsets of the data to train base models promotes more differences between the base models.

Random Forests, which “bag” decision trees, can achieve very high classification accuracy.

## Bagging’s magic decrease of model variance

One of the biggest advantages of Random Forests is that they decrease variance without increasing bias. Essentially you can get a better model without having to trade off between bias and variance.

VARIANCE DECREASE

Base model estimates are averaged together, so variability of model predictions (across hypothetical samples) is lower.

NO/LITTLE BIAS INCREASE

The bias remains the same as the bias of the individual base models. The model is still able to model the “true function” since the base models’ complexity is unrestricted (low bias).

Enough of theory now let’s dive into the implementation  logistic regression .

We will use implementation provided by the python machine learning framework known as scikit-learn.

## Problem Statement :

To build a simple Random Forest model for prediction of car quality given other attributes about the car.

## Data details

==========================================
1. Title: Car Evaluation Database==========================================
The dataset is available at  “http://archive.ics.uci.edu/ml/datasets/Car+Evaluation”
2. Sources:
(a) Creator: Marko Bohanec
(b) Donors: Marko Bohanec   (marko.bohanec@ijs.si)
Blaz Zupan      (blaz.zupan@ijs.si)
(c) Date: June, 1997

3. Past Usage:

The hierarchical decision model, from which this dataset is derived, was first presented in M. Bohanec and V. Rajkovic: Knowledge acquisition and explanation for
multi-attribute decision making. In 8th Intl Workshop on Expert Systems and their Applications, Avignon, France. pages 59-78, 1988.

Within machine-learning, this dataset was used for the evaluation of HINT (Hierarchy INduction Tool), which was proved to be able to
completely reconstruct the original hierarchical model. This,together with a comparison with C4.5, is presented in

B. Zupan, M. Bohanec, I. Bratko, J. Demsar: Machine learning by function decomposition. ICML-97, Nashville, TN. 1997 (to appear)

4. Relevant Information Paragraph:

Car Evaluation Database was derived from a simple hierarchical decision model originally developed for the demonstration of DEX
(M. Bohanec, V. Rajkovic: Expert system for decision making. Sistemica 1(1), pp. 145-157, 1990.). The model evaluates
cars according to the following concept structure:

CAR                      car acceptability
. PRICE                  overall price
. . maint                price of the maintenance
. TECH                   technical characteristics
. . COMFORT              comfort
. . . doors              number of doors
. . . persons            capacity in terms of persons to carry
. . . lug_boot           the size of luggage boot
. . safety               estimated safety of the car

Input attributes are printed in lowercase. Besides the target concept (CAR), the model includes three intermediate concepts:
PRICE, TECH, COMFORT. Every concept is in the original model related to its lower level descendants by a set of examples (for
these examples sets see http://www-ai.ijs.si/BlazZupan/car.html).

The Car Evaluation Database contains examples with the structural information removed, i.e., directly relates CAR to the six input attributes: buying, maint, doors, persons, lug_boot, safety.

Because of known underlying concept structure, this database may be particularly useful for testing constructive induction and
structure discovery methods.

5. Number of Instances: 1728
(instances completely cover the attribute space)

6. Number of Attributes: 6

7. Attribute Values:

maint        v-high, high, med, low
doors        2, 3, 4, 5-more
persons      2, 4, more
lug_boot     small, med, big
safety       low, med, high

8. Missing Attribute Values: none

9. Class Distribution (number of instances per class)

class      N          N[%]
—————————–
unacc     1210     (70.023 %)
acc        384     (22.222 %)
good        69     ( 3.993 %)
v-good      65     ( 3.762 %)

Tools to be used :
Numpy,pandas,scikit-learn

Python Implementation with code :
Import necessary libraries

Import the necessary modules from specific libraries.

```import os
import numpy as np
import pandas as pd
import numpy as np, pandas as pd
import matplotlib.pyplot as plt
from sklearn import  metrics, model_selection, preprocessing
from sklearn.ensemble import  RandomForestClassifier
```

Use pandas module to read the bike data from the file system. Check few records of the dataset.

```data = pd.read_csv('data/car_quality/car.data',names=['buying','maint','doors','persons','lug_boot','safety','class'])
```
```buying    maint       doors      persons  lug_boot safety      class

0              vhigh      vhigh      2              2              small       low          unacc

1              vhigh      vhigh      2              2              small       med         unacc

2              vhigh      vhigh      2              2              small       high        unacc

3              vhigh      vhigh      2              2              med         low          unacc

4              vhigh      vhigh      2              2              med         med         unacc

```

Check few information about the data set

```data.info()
```
```<class 'pandas.core.frame.DataFrame'>
RangeIndex: 1728 entries, 0 to 1727
Data columns (total 7 columns):
maint       1728 non-null object
doors       1728 non-null object
persons     1728 non-null object
lug_boot    1728 non-null object
safety      1728 non-null object
class       1728 non-null object
dtypes: object(7)
memory usage: 94.6+ KB
```

The train data set has 1728 rows and 7 columns.
There are no missing values in the dataset

Identify the target variable

```data['class'],class_names = pd.factorize(data['class'])
```

The target variable is marked as class in the dataframe. The values are present in  string format. However the algorithm requires the variables to be coded into its equivalent integer codes. We can convert the string categorical values into a integer code using factorize method of the pandas library.

Let’s check the encoded values now.

```print(class_names)
print(data['class'].unique())

```
```Index([u'unacc', u'acc', u'vgood', u'good'], dtype='object')
[0 1 2 3]
```

As we can see the values has been encoded into 4 different numeric labels.

Identify the predictor variables and encode any string variables to equivalent integer codes

```data['buying'],_ = pd.factorize(data['buying'])
data['maint'],_ = pd.factorize(data['maint'])
data['doors'],_ = pd.factorize(data['doors'])
data['persons'],_ = pd.factorize(data['persons'])
data['lug_boot'],_ = pd.factorize(data['lug_boot'])
data['safety'],_ = pd.factorize(data['safety'])
```
```buying            maint       doors      persons  lug_boot safety      class

0              0              0              0              0              0              0              0

1              0              0              0              0              0              1              0

2              0              0              0              0              0              2              0

3              0              0              0              0              1              0              0

4              0              0              0              0              1              1              0

```

Check the data types now :

```data.info()
```
```<class 'pandas.core.frame.DataFrame'>
RangeIndex: 1728 entries, 0 to 1727
Data columns (total 7 columns):
maint       1728 non-null int64
doors       1728 non-null int64
persons     1728 non-null int64
lug_boot    1728 non-null int64
safety      1728 non-null int64
class       1728 non-null int64
dtypes: int64(7)
memory usage: 94.6 KB

```

Everything is now converted in integer form.

Select the predictor feature and select the target variable

```X = data.iloc[:,:-1]
y = data.iloc[:,-1]

```

Train test split :

```# split data randomly into 70% training and 30% test
X_train, X_test, y_train, y_test = model_selection.train_test_split(X, y, test_size=0.3, random_state=0)
```

Training / model fitting

```# we can achieve the above two tasks using the following codes
# Bagging: using all features
model = RandomForestClassifier(random_state=1)
model.fit(X_train, y_train)
```

Model parameters study :

```# use the model to make predictions with the test data
y_pred = model.predict(X_test)
# how did our model perform?
count_misclassified = (y_test != y_pred).sum()
print('Misclassified samples: {}'.format(count_misclassified))
accuracy = metrics.accuracy_score(y_test, y_pred)
print('Accuracy: {:.2f}'.format(accuracy))
```
```Misclassified samples: 19
Accuracy: 0.96
```

As you can see the algorithm was able to achieve classification accuracy of 96% on the held out set. Only 19 samples were misclassified.

• Random Forest can be used to solve both kinds of problems ;regression and classification.
• Is capable of handling high dimensional datasets
• Can be used to extract out relevant features
• Handles missing data effectively internally