Evaluating the Performance of AlexNet and SVM for Tourism

Recommendation

T. Jaivanth Reddy

and K. Vijayalakshmi

†

Department of Electronics and Communication Engineering, Saveetha School of Engineering, Saveetha Institute of Medical

and Technical Sciences, Saveetha University, Chennai, Tamil Nadu, 602105, India

Keywords: Novel AlexNet Classifier, Support Vector Machine, Tourism Recommendation, Tourist Destination, Travel,

Technology.

Abstract: This study aimed to enhance a tourism recommendation system using the novel AlexNet classifier, contrasting

it with the Support Vector Machine (SVM) algorithm. An alpha value of 0.05 and G Power of 0.8 determined

an appropriate sample size, with a confidence interval of 95%. Of the 5,456 samples, 3,819 were for training

and 1,637 for testing. The AlexNet and SVM algorithms were labelled as "Group 1" and "Group 2",

respectively, and both underwent 20 test iterations. Results revealed the AlexNet Classifier achieved a 97.20%

accuracy rate, surpassing the SVM's 92.45%. A significant statistical difference was confirmed between the

two algorithms, suggesting AlexNet provides more accurate travel recommendations.

1 INTRODUCTION

The purpose of these systems is to assist travellers in

discovering new tourist destinations and experiences

that suit their interests and budget and to provide

personalised recommendations for planning trips

(Duen-Yian Yeh, 2015). This technology is a

valuable resource for travellers and travel-related

businesses, facilitating the discovery and planning of

travel experiences and enhancing trip enjoyment

using the innovative categorisation technique.

Tourism recommendation systems are employed in

various contexts to help travellers discover and plan

their trips. Online travel agencies such as Expedia and

Booking.com might use these systems to recommend

places and activities based on prior reservations and

interests (Kevin Meehan, 2013) (Palanivelu, J. et al.

2022). Travel agencies can harness recommendation

system technology to plan tailored holidays and

business trips. Simultaneously, tourism boards and

destination marketing organisations might promote

local attractions and activities using the innovative

categorisation technique (Aiden McCaughey, 2014).

Airlines might also adopt recommendation systems to

suggest tourist destinations and activities based on

past bookings and preferences, offering additional

Research Scholar

†

Project Guide, Corresponding Author

travel-related products and services like car hires and

hotel reservations. By using the creative

categorisation technique, tourism recommendation

systems can be beneficial for various travel-related

businesses and organisations, attracting and retaining

customers with relevant, personalised

recommendations (Ricardo Colomo-Palacios, 2017).

Over the past five years, nearly 175 articles on

tourism recommendation systems have been

published in sources such as IEEE Xplore, Google

Scholar, and Springer. These systems let users enter a

photo or a keyword detailing their desired visit type

and then scour a database for tourist destinations that

match the visual traits or keywords provided

(Liangliang Cao, 2010) (Karthik B et al. 2022). The

system categorises a vast set of geotagged web photos

by location, picking out representative images for

each group, subsequently offering these as

recommendations to users (Andrew Gallagher, 2021).

As smartphone manufacturers integrate more sensors,

developers can discern a user's context with increased

accuracy, pivoting to a multifaceted contextual

approach rather than a sole reliance on location

(Damianos Gavalas, 2014). A comprehensive review

of smart e-Tourism recommendation systems

featured in Artificial Intelligence journals and

conferences since 2008 has been undertaken.

Reddy, T. and Vijayalakshmi, K.

Evaluating the Performance of AlexNet and SVM for Tourism Recommendation.

DOI: 10.5220/0012544000003739

Paper published under CC license (CC BY-NC-ND 4.0)

In Proceedings of the 1st International Conference on Artiﬁcial Intelligence for Internet of Things: Accelerating Innovation in Industry and Consumer Electronics (AI4IoT 2023), pages 44-50

ISBN: 978-989-758-661-3

The literature review on tourism recommendation

systems presents several gaps. A notable absence is

research investigating personalisation's influence on

system efficacy. Although numerous tourism

recommendation systems profess to furnish

personalised advice, scant research delves into how

adeptly these systems grasp individual proclivities or

the consequential effect on recommendation quality.

In this project, an innovative categorisation method

classifies tourist sites based on location specialities

and user interests, bolstering accuracy. Thus, this

research's paramount objective is to heighten the

accuracy of the tourism recommendation system,

favouring the novel AlexNet classifier over the

Support Vector Machine algorithm.

2 MATERIALS AND METHODS

The purpose of these systems is to assist travellers in

discovering new tourist destinations and experiences

that suit their interests and budget and to provide

personalised recommendations for planning trips

(Duen-Yian Yeh, 2015). This technology is a

valuable resource for travellers and travel-related

businesses, facilitating the discovery and planning of

travel experiences and enhancing trip enjoyment

using the innovative categorisation technique.

Tourism recommendation systems are employed in

various contexts to help travellers discover and plan

their trips. Online travel agencies such as Expedia and

Booking.com might use these systems to recommend

places and activities based on prior reservations and

interests (Kevin Meehan, 2013) (Palanivelu, J. et al.

2022). Travel agencies can harness recommendation

system technology to plan tailored holidays and

business trips. Simultaneously, tourism boards and

destination marketing organisations might promote

local attractions and activities using the innovative

categorisation technique (Aiden McCaughey, 2014).

Airlines might also adopt recommendation systems to

suggest tourist destinations and activities based on

past bookings and preferences, offering additional

travel-related products and services like car hires and

hotel reservations. By using the creative

categorisation technique, tourism recommendation

systems can be beneficial for various travel-related

businesses and organisations, attracting and retaining

customers with relevant, personalised

recommendations (Ricardo Colomo-Palacios, 2017).

Over the past five years, nearly 175 articles on

tourism recommendation systems have been

published in sources such as IEEE Xplore, Google

Scholar, and Springer. These systems let users enter a

photo or a keyword detailing their desired visit type

and then scour a database for tourist destinations that

match the visual traits or keywords provided

(Liangliang Cao, 2010) (Karthik B et al. 2022). The

system categorises a vast set of geotagged web photos

by location, picking out representative images for

each group, subsequently offering these as

recommendations to users (Andrew Gallagher, 2021).

As smartphone manufacturers integrate more sensors,

developers can discern a user's context with increased

accuracy, pivoting to a multifaceted contextual

approach rather than a sole reliance on location

(Damianos Gavalas, 2014). A comprehensive review

of smart e-Tourism recommendation systems

featured in Artificial Intelligence journals and

conferences since 2008 has been undertaken.

The literature review on tourism recommendation

systems presents several gaps. A notable absence is

research investigating personalisation's influence on

system efficacy. Although numerous tourism

recommendation systems profess to furnish

personalised advice, scant research delves into how

adeptly these systems grasp individual proclivities or

the consequential effect on recommendation quality.

In this project, an innovative categorisation method

classifies tourist sites based on location specialities

and user interests, bolstering accuracy. Thus, this

research's paramount objective is to heighten the

accuracy of the tourism recommendation system,

favouring the novel AlexNet classifier over the

Support Vector Machine algorithm.

2.1 AlexNet Classifier

AlexNet is a classifier that utilises a deep neural

network architecture to identify patterns and features in

input data, predicting the class to which it belongs

(Priyadarshiny Dhar, 2021). What distinguished

AlexNet was the employment of the ReLU activation

function coupled with dropout regularisation

technology. These advancements substantially

enhanced the model's capability to generalise to new

data and curtailed overfitting. It has mastered the

recognition of an extensive array of image features and

can classify new images based on these learned

attributes.

Pseudo code

Input: An image of size 227 x 227 x 3

Output: The predicted class label

# Define the AlexNet architecture

1. Convolution layer 1 with 96 filters of size

11x11, stride 4, and padding 0, with ReLU

activation

Evaluating the Performance of AlexNet and SVM for Tourism Recommendation

2. Max pooling layer with kernel size 3x3 and

stride 2

3. Fully connected layer with 4096 neurons and

ReLU activation

4. Dropout layer with a probability of 0.5

5. Fully connected layer with 4096 neurons and

ReLU activation

6. Dropout layer with a probability of 0.5

7. Output layer with 1000 neurons

(corresponding to the 1000 ImageNet classes)

and softmax activation

# Preprocess the input image

1. Subtract the mean RGB values of the training

set from the input image

2. Scale the pixel values to [0, 1]

# Forward pass through the network

1. Pass the preprocessed input image through the

convolutional layers, pooling layers, and fully

connected layers

2. Compute the softmax probabilities for each

class using the output layer

# Return the predicted class label

1. Retrieve the class identifier associated with

the maximum softmax probability

2.2 Support Vector Machine (SVM)

Algorithm

Support Vector Machine (SVM) is a popular

supervised learning algorithm used for classification

and regression tasks. It is especially effective for

classification problems. The objective of the SVM

algorithm is to create the optimal decision boundary,

termed a hyperplane, that can divide an n-

dimensional space into classes to classify new data

points accurately (David L. Olson, Dursun Delen,

2017). The SVM algorithm selects the extreme

points, termed support vectors, that assist in creating

the hyperplane. These support vectors lend the

algorithm its name: Support Vector Machine.

The testing was conducted using a Jupyter

Notebook on a hardware device equipped with an

AMD Ryzen 5 3500U processor, 8GB of RAM, a

1TB HDD, and a 256GB SSD, running the Windows

11 operating system. Both the Chrome browser and

SPSS software were utilised for statistical analysis.

2.3 Testing Procedure

To perform semi-supervised clustering on a dataset

using the Anaconda Navigator and Jupyter Notebook,

follow these steps:

1. Install the Anaconda Navigator and launch it.

2. Open a Jupyter Notebook by entering the

command "jupyter notebook" in the terminal.

3. Create a new notebook by clicking the "New"

button in the top right corner.

4. In the notebook's first cell, install and import

the necessary libraries: pandas, numpy,

matplotlib, seaborn, sklearn, tensorflow, and

jupyter themes.

5. Load the dataset – a CSV file from Github

with 5456 records – which will be split for

training and testing in a 70:30 ratio.

6. Divide the dataset into separate sets for testing

and training.

7. Input the Python code to execute semi-

supervised clustering in a cell.

8. Execute the code by clicking the "Run" button.

9. Note the model's accuracy in an Excel sheet

and further analyse it using SPSS software.

2.4 Statistical Analysis

In this research study, IBM SPSS Version 26 is used

for an exhaustive statistical analysis of multiple

variables. The main aim is to assess the mean

accuracy using the Independent Sample T-Test. The

study also employs bivariate correlation analysis in

SPSS to produce a detailed correlation table

(Okagbue 2021). The independent variables

examined are "places" and "reviews", with

"accuracy" being the dependent variable.

The analysis considers independent variables such

as accuracy, standard mean error, and standard

deviation (Okagbue 2021). An Independent Sample

T-Test is carried out on these variables to scrutinise

the outcomes. The dependent variables in focus are

the AlexNet Classifier and the SVM algorithm.

3 RESULTS

Tourist destinations can be more effectively

recommended using these systems. The independent

sample T-test compared the accuracy between the

AlexNet Classifier and SVM algorithms. The results

demonstrated that the AlexNet Classifier achieved a

higher accuracy rate of 97.20% compared to the SVM

algorithm's 92.45%. The p-value of 0.000 from the

AI4IoT 2023 - First International Conference on Artiﬁcial Intelligence for Internet of things (AI4IOT): Accelerating Innovation in Industry

and Consumer Electronics

independent sample T-test signifies a statistically

significant difference between the two algorithms.

Table 1 displays the mean accuracy, standard

deviation, and standard error mean for both the

AlexNet Classifier and the Support Vector Machine

algorithm. The AlexNet Classifier's average accuracy

is 97.20%, while the Support Vector Machine

algorithm records a mean accuracy of 92.45%. Table

2 offers a comparative review of raw data values for

both algorithms. This analysis uses a dataset of 40

samples, split evenly with 20 samples for each

algorithm. Table 3 presents the independent sample

T-Test results for the AlexNet Classifier and Support

Vector Machine algorithm, detailing the significance

and standard error. The study assumed equal

variances of 4.531.

Figure 1 showcases the comparative mean

accuracy between the AlexNet Classifier and the

SVM algorithm. It's evident from the figure that the

AlexNet Classifier, with a mean accuracy of 97.20%,

outperforms the Support Vector Machine algorithm,

which has a mean accuracy of 92.45%.

Table 1: The SVM algorithm's mean accuracy is 92.4585, compared to 97.2080 for the AlexNet Classifier. Additionally, the

following table reveals that the AlexNet standard deviation is 2.10600 and the standard error mean is 0.47092.

Group

Statistics

Algorithm

Mean

Std. Deviation

Std. Error Mean

Accuracy

AlexNet

97.2080

2.10600

0.47092

SVM

92.4585

4.18753

0.93636

Table 2: Accuracy of AlexNet and SVM of 20 samples each. AlexNet Classifier has given the highest accuracy of 99.20 and

the SVM algorithm has given the accuracy of 98.89.

SAMPLES

GROUP 1Accuracy in %

(AlexNet)

GROUP 2 Accuracy in % (SVM)

TEST 1

98.05

82.78

TEST 2

97.65

88.62

TEST 3

97.02

88.74

TEST 4

99.86

89.98

TEST 5

96.12

88.56

TEST 6

96.45

87.28

TEST 7

98.66

89.99

TEST 8

97.55

90.12

TEST 9

98.87

98.89

TEST 10

98.45

92.56

TEST 11

98.22

95.63

TEST 12

96.67

96.12

TEST 13

96.45

95.41

TEST 14

97.52

96.09

TEST 15

98.75

94.51

TEST 16

91.20

98.80

TEST 17

92.80

95.66

TEST 18

98.77

93.62

TEST 19

95.90

98.06

TEST 20

99.20

97.40

Evaluating the Performance of AlexNet and SVM for Tourism Recommendation

Table 3: An independent sample T-Test analysis of the AlexNet Classifier and Support Vector Machine algorithm, with the

significance value of 0.000 and standard error of 1.04811.

Independent samples

test

Levene’s Test for

Equality of

Variances

t-test for Equality of Means

Sig

Sig(2-

tailed)

Mean

Differenc

Std.

Error

Difference

95% Confidence Interval

of the Difference

Lower

Upper

accuracy

Equal

variances

assumed

10.425

0.003

4.531

0.000

4.74950

1.04811

2.62771

6.87129

Equal

variances

not

assumed

4.531

28.033

0.000

4.74950

1.04811

2.60266

6.89634

Figure 1: Analysis of the AlexNet and SVM Classifier. The AlexNet and SVM have respective mean accuracy of 97.20% and

92.45%. X axis: Alex Net Classifier vs SVM Classifier, Y axis: Mean Accuracy +/- 1 SD.

4 DISCUSSION

The AlexNet Classifier algorithm demonstrated

superior accuracy for tourism recommendations,

recording an accuracy rate of 97.20% compared to the

SVM algorithm's 92.45%. Statistically, the findings

were significant, evidenced by a p-value of 0.000.

This signals a notable difference between the two

algorithms' performance. The adoption of the

AlexNet Classifier markedly elevated the SVM

algorithm's accuracy, as highlighted by a p-value

below 0.05.

The majority of researchers and industry experts

concur that tourism recommendation systems offer

invaluable support to travellers in uncovering new

tourist destinations and experiences using innovative

categorisation techniques. These systems expedite the

recommendation process, aligning tourists with

destinations and activities tailored to their interests

and budgets. They are particularly beneficial for

travellers pressed for time or those seeking bespoke

travel experiences (Eleonora Pantano, 2019). The

systems also generate personalised recommendations

rooted in travellers' past activities and inclinations,

helping unveil destinations or experiences previously

AI4IoT 2023 - First International Conference on Artiﬁcial Intelligence for Internet of things (AI4IOT): Accelerating Innovation in Industry

and Consumer Electronics

overlooked (Santamaria-Granados, Mendoza-

Moreno, and Ramirez-Gonzalez 2020). Moreover,

tourism recommendation platforms assist businesses

like hotels and tour providers in effectively pairing

travellers with appealing destinations or activities

(Mohamed Elyes Ben Haj Kbaier, 2018).

Collectively, these systems serve as a potent tool for

both travellers and travel-related businesses, fostering

effortless travel experience discovery and crafting

enriching journeys (Torres-Ruiz, 2018). Nonetheless,

some experts argue that these systems might not

consistently yield accurate or trustworthy

recommendations (Petrevska and Koceski, 2012).

This scepticism stems from the system's reliance on

data and algorithms that might not holistically

represent intricate individual preferences, sometimes

achieving accuracy as low as 79% (Shafqat and Byun

2019). Traditional tour planning systems generally

adopt a tripartite structure: delineating tourist

profiles, assessing Points of Interest (POIs), and route

optimisation. Parallel research showcases a

methodology that permits tourists to define their

interests via image collections, facilitating the

system's deduction of their profile. Following the

user's choices, the system consistently amends their

dynamic profile, reaching accuracy levels of 78.4%

(Konstantinos Pliakos, 2015). The system

subsequently curates a resource list, boasting 90%

accuracy, harmonised with both the user's profile and

destination tourist resources (Linaza 2011).

Nonetheless, there's potential system bias towards

specific destinations or activities, which may stem

from user demographic data or underlying

algorithmic biases (Mehrbakhsh Nilashi, 2017).

Tourism recommendation systems are not devoid

of challenges, including noise, erroneous or

unsuitable data. Effective recommendation systems

often necessitate copious user data. In data scarcity,

the recommendations might be imprecise. Future

potential for tourism recommendation systems lies in

furnishing personalised suggestions, interoperability

with various systems, synergy with social media, and

mobile-centric optimisation through innovative

categorisation techniques.

5 CONCLUSION

In conclusion, several salient points have emerged

from the examination of the tourism recommendation

system and the comparison of the AlexNet Classifier

with the SVM algorithm:

● The AlexNet Classifier has proven to be more

effective in tourism recommendations,

achieving a superior accuracy rate of 97.20%.

● The SVM algorithm, while still effective,

lagged behind with an accuracy rate of

92.45%.

● There is a broad consensus among researchers

and industry practitioners that tourism

recommendation systems, harnessing

innovative categorisation techniques,

significantly enhance the traveller experience

by providing tailored suggestions.

● Personalised recommendations, derived from

past behaviours and preferences, enable

travellers to discover novel destinations and

activities that might otherwise be overlooked.

● There's some caution within the industry, with

concerns regarding the potential inaccuracies

of recommendation systems, especially when

there is insufficient or noisy data.

● Future prospects for tourism recommendation

systems include their integration with social

media platforms, optimisation for mobile use,

and their potential to deliver increasingly

tailored recommendations to users.

REFERENCES

Changjie Tang, Charles X. Ling (2007), Advanced Data

Mining and Applications. Springer Berlin Heidelberg.

Mohamed Elyes Ben Haj Kbaier (2017), “A Personalized

Hybrid Tourism Recommender System.”

Mehrbakhsh Nilashi, Karamollah Bagherifard (2017). “A

Recommender System for Tourism Industry Using

Cluster Ensemble and Prediction Machine Learning

Techniques.”

Liangliang Cao; Jiebo Luo, (2010) “A worldwide Tourism

Recommendation System Based on Geotaggedweb

Photos.”

Karthik B and Ramkumar G (2022)" Comparison of

Feature Extraction Technique for Segmentation in

Human Iris Recognition Under Uncontrolled

Environment Using CNN Algorithm with SVM

Classifier" ECS Transactions, doi:

10.1149/10701.16785ecst

Kevin Meehan; Tom Lunney (2013), Context-Aware

Intelligent Recommendation System for Tourism.

Priyadarshiny Dhar, Saibal Dutta (2021)“Cross-Wavelet

Assisted Convolution Neural Network (AlexNet)

Approach for Phonocardiogram Signals

Classification.”

Eleonora Pantano, Constantinos-Vasilios Priporas,

(2019)“Facilitating Tourists’ Decision Making through

Open Data Analyses: A Novel Recommender System.”

Rula A. Hamid, A.S. Albahri, (2021). “How Smart Is E-

Tourism? A Systematic Review of Smart Tourism

Evaluating the Performance of AlexNet and SVM for Tourism Recommendation

Recommendation System Applying Data

Management.”

Joan Borràs a b, Antonio Moreno (2014) “Intelligent

Tourism Recommender Systems: A Survey.”

Linaza, Maria Teresa, Amaia Agirregoikoa, Ander Garcia,

Jose Ignacio Torres, and Kepa Aranburu, (2011).

“Image-Based Travel Recommender System for Small

Tourist Destinations.”

Damianos Gavalas a b, Charalampos Konstantopoulos

(2014). “Mobile Recommender Systems in Tourism.”

Journal of Network and Computer Applications 39

(March): 319–33.

Konstantinos Pliakos and Constantine Kotropoulos (2015)

Accessed December 30, 2022.

https://doi.org/10.1142/S0218213015400217.

Okagbue, Hilary I., Pelumi E. Oguntunde, Emmanuela C.

M. Obasi, and Elvir M. Akhmetshin. (2021). “Trends

and Usage Pattern of SPSS and Minitab Software in

Scientific Research.” Journal of Physics. Conference

Series 1734 (1): 012017.

Palanivelu, J., Thanigaivel, S., Vickram, S., Dey, N.,

Mihaylova, D., & Desseva, I. (2022). Probiotics in

functional foods: survival assessment and approaches

for improved viability. Applied Sciences, 12(1), 455.

Petrevska, Biljana, and Saso Koceski. (2012). “Tourism

Recommendation System: Empirical Investigation.”

Duen-Yian Yeh, Ching-Hsue Cheng (2015),

“Recommendation System for Popular Tourist

Attractions in Taiwan Using Delphi Panel and

Repertory Grid Techniques.”

Santamaria-Granados, Luz, Juan Francisco Mendoza-

Moreno, Angela Chantre-Astaiza, Mario Munoz-

Organero, and Gustavo Ramirez-Gonzalez. (2021).

“Tourist Experiences Recommender System Based on

Emotion Recognition with Wearable Data.”

Santamaria-Granados, Luz, Juan Francisco Mendoza-

Moreno, and Gustavo Ramirez-Gonzalez. (2020).

“Tourist Recommender Systems Based on Emotion

Recognition—A Scientometric Review.”

Shafqat, Wafa, and Yung-Cheol Byun. (2019). “A

Recommendation Mechanism for Under-Emphasized

Tourist Spots Using Topic Modeling and Sentiment

Analysis.”

Torres-Ruiz, Miguel, Felix Mata, Roberto Zagal, Giovanni

Guzmán, Rolando Quintero, and Marco Moreno-Ibarra.

(2018). “A Recommender System to Generate Museum

Itineraries Applying Augmented Reality and Social-

Sensor Mining Techniques.”

Ricardo Colomo-Palacios, Francisco José García-Peñalvo

(2017). “Towards a Social and Context-Aware Mobile

Recommendation System for Tourism.” 2017.

Pervasive and Mobile Computing 38 (July): 505–15.

AI4IoT 2023 - First International Conference on Artiﬁcial Intelligence for Internet of things (AI4IOT): Accelerating Innovation in Industry

and Consumer Electronics