ML-Supervised-Prediction-of-Driving-Conditions

The aim of this project was to predict driving style, road conditions or traffic state based on sensor inputs directly from a vehicle. Two data-sets were used, an Opel Corsa & a Peugeot 207.

import pandas as pd
import numpy as np
import sklearn
from sklearn.metrics import confusion_matrix, accuracy_score, balanced_accuracy_score, f1_score
from sklearn.preprocessing import OrdinalEncoder, OneHotEncoder
from sklearn.model_selection import train_test_split

Data Exploration

#Download and concatenation of data
data11 = pd.read_csv('opel_corsa_01.csv', delimiter=';')
data12 = pd.read_csv('opel_corsa_02.csv', delimiter=';')
data21 = pd.read_csv('peugeot_207_01.csv', delimiter=';')
data22 = pd.read_csv('peugeot_207_02.csv', delimiter=';')
data   = pd.concat([data11, data12, data21, data22], ignore_index=True)

data

	AltitudeVariation	VehicleSpeedInstantaneous	VehicleSpeedAverage	VehicleSpeedVariance	VehicleSpeedVariation	LongitudinalAcceleration	EngineLoad	EngineCoolantTemperature	ManifoldAbsolutePressure	EngineRPM	MassAirFlow	IntakeAirTemperature	VerticalAcceleration	FuelConsumptionAverage	roadSurface	traffic	drivingStyle
0	NaN	0.000000	NaN	NaN	NaN	0.0156	25.490196	64.0	100.0	801.0	7.850000	22.0	-0.0078	NaN	SmoothCondition	LowCongestionCondition	EvenPaceStyle
1	NaN	0.000000	NaN	NaN	NaN	0.0156	25.490196	64.0	100.0	803.0	7.890000	22.0	-0.0156	NaN	SmoothCondition	LowCongestionCondition	EvenPaceStyle
2	NaN	NaN	NaN	NaN	NaN	0.0273	25.882353	64.0	100.0	800.0	7.770000	22.0	-0.0273	NaN	SmoothCondition	LowCongestionCondition	EvenPaceStyle
3	NaN	NaN	NaN	NaN	NaN	0.0391	25.882353	64.0	100.0	798.0	7.770000	22.0	-0.0273	NaN	SmoothCondition	LowCongestionCondition	EvenPaceStyle
4	NaN	0.000000	NaN	NaN	NaN	0.0469	25.882353	65.0	100.0	798.0	7.940000	22.0	-0.0312	NaN	SmoothCondition	LowCongestionCondition	EvenPaceStyle
...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...
24952	1.000000	28.799999	28.559999	57.190571	3.600000	-0.0292	25.882353	81.0	115.0	1755.5	20.469999	25.0	-0.1661	14.578003	SmoothCondition	LowCongestionCondition	EvenPaceStyle
24953	1.699997	30.599998	28.529999	57.010266	1.799999	-0.0304	11.764706	81.0	106.0	736.5	17.740000	25.0	-0.1987	14.585642	SmoothCondition	LowCongestionCondition	EvenPaceStyle
24954	1.800003	29.699999	28.499999	56.883045	-0.900000	-0.1684	98.039215	81.0	106.0	1254.0	9.520000	24.0	-0.1156	14.547294	SmoothCondition	LowCongestionCondition	EvenPaceStyle
24955	2.100006	29.699999	28.409999	56.160910	0.000000	-0.0644	79.607841	80.0	112.0	1254.0	14.910000	23.0	-0.0760	14.546828	SmoothCondition	LowCongestionCondition	EvenPaceStyle
24956	1.500000	33.299999	28.349999	55.340843	3.600000	-0.1817	80.000000	80.0	113.0	1363.5	15.330000	23.0	-0.0605	14.554068	SmoothCondition	LowCongestionCondition	EvenPaceStyle

counts_rs = data['roadSurface'].value_counts(sort=False)

total_rs = 0
for entry in counts_rs:
    total_rs += entry

for i in range(len(counts_rs)):
    print(counts_rs.index[i] , ':' , round((counts_rs[i]/total_rs)*100,2), '%' )

SmoothCondition : 61.07 %
UnevenCondition : 25.91 %
FullOfHolesCondition : 13.02 %

counts_t = data['traffic'].value_counts(sort=False)

total_t = 0
for entry in counts_t:
    total_t += entry

for i in range(len(counts_t)):
    print(counts_t.index[i] , ':' , round((counts_t[i]/total_t)*100,2), '%' )

LowCongestionCondition : 75.21 %
NormalCongestionCondition : 12.71 %
HighCongestionCondition : 12.09 %

counts_ds = data['drivingStyle'].value_counts(sort=False)

total_ds = 0
for entry in counts_ds:
    total_ds += entry

for i in range(len(counts_ds)):
    print(counts_ds.index[i] , ':' , round((counts_ds[i]/total_ds)*100,2), '%' )

EvenPaceStyle : 88.51 %
AggressiveStyle : 11.49 %

Data Prep

#Input data
x = data.values[::, 0:14]

#NaN, Standartization, Label Encoder
from sklearn.impute import SimpleImputer
imp = SimpleImputer(strategy="most_frequent")
x = imp.fit_transform(x)

from sklearn.preprocessing import StandardScaler
scaler = StandardScaler().fit(x)
x = scaler.fit_transform(x)

def processYandSplit(range):
    #Feature Assignment
    y = data.values[::, range]

    enc = OrdinalEncoder()
    # enc = OneHotEncoder()
    y = enc.fit_transform(y)

    #Split into training & test data with a ratio of 80:20
    x_train, x_test, y_train, y_test = train_test_split(x, y, test_size=0.2, random_state=0)

    # Make y_train 1 dimensional
    y_train = y_train.ravel()
    return x_train, x_test, y_train, y_test

Models

def computeSVC(x_train, x_test, y_train, y_test):
    #Train the SVC model
    from sklearn.svm import SVC
    svc = SVC(kernel='linear')
    model = svc.fit(x_train, y_train)

    #Predict the test data
    resultSVC = svc.predict(x_test)

    # Assess the success with a confusion matrix
    conf = confusion_matrix(y_test, resultSVC)
    pd.DataFrame(conf, index=svc.classes_, columns=svc.classes_)
    
    return resultSVC, conf

def computeLR(x_train, x_test, y_train, y_test, solver='lbfgs', penalty='l2', C=0.1):
    from sklearn.linear_model import LogisticRegression
    #Train the LR model
    clf = LogisticRegression(random_state=0, solver=solver, penalty=penalty, C=C, multi_class='multinomial')

    model2 = clf.fit(x_train, y_train)
    #Predict the test data
    resultLR = clf.predict(x_test)

    # Assess the success with a confusion matrix
    conf = confusion_matrix(y_test, resultLR)
    pd.DataFrame(conf, index=clf.classes_, columns=clf.classes_)
    
    return resultLR, conf

Results

results_df = pd.DataFrame({})
# For each target feature:
for i in [14, 15, 16]:
    # Split data sets
    x_train, x_test, y_train, y_test = processYandSplit(slice(i,i+1))
    # Compute SVC
    resultSVC, confSVC = computeSVC(x_train, x_test, y_train, y_test)
    # Compute Logistic Regression 
    resultLR, confLR   = computeLR(x_train, x_test, y_train, y_test)
    # Build a results DataFrame
    accuracy_data = [[accuracy_score(y_test, resultSVC), balanced_accuracy_score(y_test, resultSVC), f1_score(y_test, resultSVC, average='macro'), accuracy_score(y_test, resultLR), balanced_accuracy_score(y_test, resultLR), f1_score(y_test, resultLR, average='macro')]]
    feature_results =  pd.DataFrame(accuracy_data, index = [data.columns[i]], columns=["SVC Acc", "SVC Bal Acc", "SVC F1 Mean", "LR Acc", "LR Bal Acc", "LR F1 Mean"])
    results_df = pd.concat([results_df, feature_results])

results_df

	SVC Acc	SVC Bal Acc	SVC F1 Mean	LR Acc	LR Bal Acc	LR F1 Mean
roadSurface	0.782652	0.737606	0.729651	0.778446	0.715456	0.719583
traffic	0.775441	0.468239	0.455450	0.782051	0.487309	0.499387
drivingStyle	0.881611	0.500000	0.468540	0.880409	0.512502	0.496514