Analysis of the Human Barriers to Using Bicycles as a Means of Transportation in Developing Cities

Abstract

In the context of mounting mobility issues in Latin American cities, bicycles are emerging as a vital sustainable solution. However, their widespread adoption is hindered by various obstacles. This study aimed to identify and prioritize human factors inhibiting bicycle use in Colombia to support the development of effective public policies, given that research in this area mainly focuses on designing and developing road infrastructure for cyclists. An artificial intelligence classification methodology was applied to data from a self-administered online survey of 2068 participants. An objective variable was constructed to classify respondents as “potential users” or “non-potential users,” and three models (Logistic Regression, Random Forest, and XGBoost) were used to analyze the predictive power of different barriers. The results from the three models consistently show that personal, convenience, and safety perception barriers are significantly more important predictors than infrastructure factors. Specifically, inconvenience due to subsequent activities, perceived insecurity when cycling, and concern about sweating were consistently ranked as the most critical barriers. Therefore, to effectively promote cycling, public policies should address not only infrastructure development but also the mitigation of subjective and logistical barriers. Thus, these results can inform the design of more holistic mobility programs and serve as a foundation for future research on sustainable mobility.

1. Introduction

In recent decades, Latin American cities have experienced rapid urban growth accompanied by mounting mobility challenges. This has solidified a transportation model dependent on motor vehicles, with negative externalities that have become unsustainable. Chronic traffic congestion, high levels of air and noise pollution, high economic costs for users and governments, and profound social inequality in access to urban opportunities are symptoms of a system in desperate need of transformation. In this context, bicycles have emerged as an increasingly relevant means of transportation for people’s daily movements and, in some cases, for the movement of goods and services. Furthermore, adopting the bicycle as a means of transportation represents an opportunity to democratize road space, reduce pressure on mass public transportation systems, and promote more humane and resilient cities [1].

This recognition is consistent with global frameworks such as the 2030 Agenda for Sustainable Development, which emphasizes active and sustainable transportation as a vital component of creating more resilient and livable cities. Bicycle use is estimated to be directly related to the fulfillment of 11 of the 17 Sustainable Development Goals [2,3].

The increased presence of bicycles in Latin American cities is no coincidence; it is the result of several factors, including greater environmental awareness, the search for more affordable transportation options, the promotion of healthier lifestyles, and the implementation of public policies in some cases [4]. While the trend of using cycling as a means of daily transportation has recently gained momentum in Latin America, it has been observed for many years in Western European countries. In these regions, cycling has evolved from being associated solely with recreation and tourism to becoming an important part of daily commutes to work, school, and entertainment. Bicycles are an excellent alternative to motorized vehicles, which produce exhaust fumes and contribute to traffic congestion. In addition to being environmentally friendly and compact, bicycles offer competitive travel times compared to private cars or public transportation, making them an increasingly viable option for citizens [5]

It is important to note that the gap between the theoretical potential of cycling and its practical implementation is defined by a complex web of barriers that extend beyond mere political will. These barriers include a lack of safe, connected, and coherent cycling infrastructure; high levels of insecurity on the roads (e.g., accidents involving motor vehicles) and in the city (e.g., risk of theft); deep-rooted sociocultural barriers that associate bicycles with low socioeconomic status or impose gender roles; and economic barriers to initial access to the vehicle. Together, these barriers create an adverse ecosystem for potential cyclists. Therefore, a thorough, structured analysis of this phenomenon is necessary to understand its multiple dimensions—from its benefits to the obstacles hindering its full development [6,7]. Significant barriers to cycling remain despite existing policies, highlighting a disconnect between policy and implementation. Current strategies often overlook the complex personal and logistical challenges people face, a bias also seen in the academic literature, which typically prioritizes infrastructure above human factors [8,9].

The international and national recognition of cycling as a viable and desirable urban solution opens a window of opportunity for effective public policy formulation and implementation [10,11]. However, the success of these policies hinges on a thorough understanding of the local contexts of the perceived and real advantages and persistent barriers. The existence of numerous policy documents and strategies developed by government authorities in cities such as Buenos Aires [12], Argentina; Bogotá [13]; and Bucaramanga, Colombia [14], alongside the persistence of significant obstacles, suggests a possible gap between policy formulation and implementation. It also suggests that the multifaceted and deeply rooted nature of the challenges may not be fully addressed by current policies. This document aims to bridge this gap by gaining firsthand insight into the perception of potential bicycle users. This study is structured into several sections to address this gap. Section 2 details the methodology, including the study design, the population studied, and how the target variable was constructed. Section 3 presents the results of the predictive models and discusses the main identified barriers, focusing on personal, logistical, and safety factors. Finally, Section 4 presents the study’s key conclusions and their implications for designing public policies that promote bicycle mobility more effectively and sustainably.

2. Materials and Methods

2.1. Study Design

This quantitative, descriptive, cross-sectional study identified and analyzed the barriers, perceptions, and factors influencing the use of bicycles for transportation in Colombia. The study used a self-administered online survey distributed through digital channels.

2.2. Population and Sample

The target population for the study was Colombian residents with diverse sociodemographic profiles (the Colombian population, according to DANE (Departamento Administrativo Nacional de Estadística-National Administrative Department of Statistics), is 48,258,494 [15]). The sample size was calculated using the formula developed by Cochran [16,17]:

$$n={\displaystyle \frac{Z^{2}p(1-p)}{E^2}}$$

(1)

where n is the required sample size, Z is the z-value associated with the desired confidence level (1.96 for 95% confidence), p is the estimated population proportion ($p=0.5$), and E is the margin of error or desired precision (5%).

A non-probability convenience sample was used, yielding a total of 2077 responses. The final sample consisted of 2068 participants who provided digital informed consent and completed the survey. Nine individuals who selected “I do not agree to participate in this survey” were excluded from the analysis. Since the sample was non-probability and convenience-based, and the survey was distributed online, most of the participants were young (mean age: 29.56 years), urban (90%), and highly educated (mostly university students), possibly because this population is more likely to be online and willing to participate in these types of surveys. Although records of bicycle users in Colombia are limited, some government entities and universities report that bicycle users in Bogotá (Colombia’s capital) are between 18 and 35 years old [18,19]. This is consistent with the results of the study.

2.3. Data Collection Instrument

The survey, titled “People’s Perception of Bicycle Mobility,” was designed and administered on the Microsoft Forms platform. The 39-question survey was structured into three main sections.

• Informed Consent: The first section included the title of the study, the researchers responsible for it, its objective, its procedure, and its estimated duration of 10 to 15 min. It also included a confidentiality statement. Participation was conditional upon explicit acceptance of the informed consent form.
• Sociodemographic Characterization: This section collected data on gender, age, level of education, type of population (urban or rural), socioeconomic stratification, and geographic location (department and city of current residence).
• Barriers and Perception: This was the focus of the instrument, which assessed participants’ perception of various factors. Topics addressed included the following:
  • Infrastructure: Quality and availability of bike lanes, safety at intersections, condition of roads, and parking facilities.
  • Environmental and Surrounding Conditions: Distance of routes, slopes, weather conditions, odors, exhaust fumes, and poor public lighting.
  • Road Safety and Coexistence: The perception of excessive traffic and vehicle speed, as well as a lack of respect from drivers toward cyclists.
  • Personal and convenience factors: Access to a bicycle; comfort compared to other modes of transportation; physical fitness; minor mechanical problems; appropriate clothing; and perception of personal safety while riding.
  • Psychosocial Factors: Fear of being attacked; perception of cycling as less sociable or enjoyable than other activities; and concerns about personal appearance (e.g., hairstyle).
  • Parental Influence (Only for Minors): The final section contained three dichotomous (yes/no) questions that explored parents’ or guardians’ perception of the minor’s ability and safety when riding a bicycle.

Most of the questions in this section used a Likert-type scale, which provides respondents with a set of options ranging from “Strongly Agree” to “Neutral.”

2.4. Data Collection Procedure

The survey was distributed from 7 January to 15 May 2025. Participants accessed the survey via an online link. The first mandatory step was to read and accept the informed consent form. After providing consent, participants proceeded to answer the questions in the following sections. Data submission was recorded via the Microsoft Forms platform.

2.5. Data Analysis Methodology

An artificial intelligence (AI) classification methodology was used to structure the data analysis and predict the probability of a person being a potential bicycle user. The process was divided into two stages:

2.5.1. Stage I

Definition and Construction of the Target Variable

Since the dataset did not include an explicit variable for bicycle use, a binary target variable called “target_bike_use” was created. This variable classifies individuals as either “potential users” or “non-potential users” of bicycles. Six key variables from the survey that represent perceived barriers to cycling were identified for its construction.

• I don’t feel safe riding a bike.
• I don’t have access to a bike.
• I’m not fit enough.
• I would get too hot and sweat a lot.
• It’s not convenient because of my other activities.
• Drivers don’t respect cyclists.

Each of these variables was considered an “active barrier” if the original Likert response was equal to or greater than 4 (“Agree” or “Strongly Agree”) on a scale of 1 to 5, where 1 is “Strongly Disagree” and 5 is “Strongly Agree.”

The classification rule for the “target_bike_use” variable was as follows:

• target_bike_use = 0 (not a potential user): A person reported three or more of these barriers as active.
• target_bike_use = 1 (potential user): If a person reported fewer than three active barriers.

Thus, the definition of the variable is based on a threshold of perceived barriers grounded in three pillars: (1) conceptual justification, (2) preliminary empirical validation, and (3) robustness checks. This classification rule aligns with models such as the Theory of Planned Behavior (TPB) [20] and the Health Belief Model (HBM) [21]. Both models postulate that perceived barriers critically determine intention and behavior. A single barrier may not deter an individual, but multiple barriers have an additive or synergistic effect. Therefore, when the number of significant barriers reaches a “tipping point,” the probability of adopting a behavior (cycling, in this case) decreases dramatically. To ensure that the threshold was not arbitrary, a sensitivity analysis exploring different cutoff points (≥2, ≥3, and ≥4 barriers) was performed (

Table 1. Comparative evaluation of model performance (Logistic Regression, Random Forest, and XGBoost).

Threshold (No. of Active Barriers)	% Non-Potential User (Target = 0)	% Potential User (Target = 1)	Interpretation
≥2	83%	17%	Too inclusive: It creates a severe class imbalance and could misclassify individuals with few barriers as “non-potential.”
≥3 (Elected)	68%	32%	Methodological balance: Identify a majority group that faces substantial barriers while ensuring that the “potential” group is large enough for robust modeling.
≥4	51%	49%	Too strict: Although it balances the classes, it could be so demanding that it misclassifies individuals with three significant barriers as “potential,” which loses specificity.

).

The optimal combination of hyperparameters for each model was determined using the GridSearchCV technique from the Scikit-learn library (

Table 2. Model hyperparameters and values for grid search.

Model	Parameter	Default Value	Grid Search Values
Logistic Regression	penalty	l2	[‘l1’, ‘l2’]
	C	1.0	[0.1, 1, 10, 100]
	solver	lbfgs	[‘liblinear’, ‘saga’]
Random Forest	n_estimators	100	[100, 200, 500]
	max_depth	None	[10, 20, 30, None]
	min_samples_split	2	[2, 5, 10]
	min_samples_leaf	1	[1, 2, 4]
	max_features	‘sqrt’	[‘sqrt’, ‘log2’]
XGBoost	learning_rate	0.3	[0.05, 0.1, 0.2]
	n_estimators	100	[100, 200, 500]
	max_depth	6	[3, 5, 7]
	subsample	1.0	[0.8, 1.0]
	colsample_bytree	1.0	[0.8, 1.0]

). The process identified the following combinations:

• Logistic Regression: The best performance was achieved with 11-type regularization (which also helps with variable selection), a regularization strength of C = 0.1, and the Saga optimizer.
• Random Forest: The optimal model consists of a forest of 500 trees (n_estimators), where each tree has a maximum depth of 20 levels (max_depth). Specific rules are also applied regarding how and when branches are split to control overfitting (min_samples_split, min_samples_leaf).
• XGBoost: The winning configuration is a slow-learning model (learning_rate: 0.05), composed of 500 very simple trees (n_estimators) (maximum depth of 3). It also uses data subsets (subsample and colsample_bytree), a key technique to ensure the model generalizes well and does not memorize the data.

In addition to careful threshold selection, residual class imbalance was directly managed during training of the more advanced models to ensure that the minority class (“potential users”) was not overlooked. Specifically, for the Random Forest model, the hyperparameter class_weight=‘balanced’ was used, which automatically adjusts class weights inversely proportional to their frequency, while for the XGBoost model, the hyperparameter scale_pos_weight was used, which was calculated to give greater weight to errors in the classification of the minority class.

A sensitivity analysis was performed to validate the robustness of the ≥3 barrier threshold and to balance theoretical validity with empirical feasibility and avoid class imbalance. The resulting distribution (68/32) ensures that machine learning models are trained more robustly. This enables direct and transparent data handling while avoiding the complex sampling techniques and potential biases that could be introduced by a more extreme imbalance, such as the 83/17 ratio observed at the ≥2 threshold.

Figure 1 shows the performance of the three models with the 68/32 distribution. The XGBoost and Random Forest models (Figure 1a,b) show robust generalization as their training (red) and validation (green) curves stay close together while converging to a ROC–AUC (Receiver Operating Characteristic–Area Under the Curve) score near 1.0, indicating no significant overfitting. The Logistic Regression model (Figure 1c) also performs well; its initial performance gap narrows as the validation curve rises with more data, confirming that this model also avoids severe overfitting [22,23,24]. So, ensemble and boosting models like XGBoost and Random Forest are superior, achieving near-perfect discrimination between potential and non-potential users. While Logistic Regression is a viable option, these more advanced models more effectively capture complex, nonlinear interactions between variables, which gives them superior predictive power. This is how these models are validated for understanding the barriers that impede cycling.

2.5.2. Stage II

Data Preprocessing: Data scrubbing was performed to improve the quality of the data by removing blank spaces and standardizing the Likert-type responses to standardize the data. Likert-type variables were coded ordinally by assigning numerical values from 1 to 5. “Strongly disagree” or “Totally disagree” was coded as 1; “Disagree” or “Disagree” as 2; “Neutral” or “Neutral” as 3; “Agree” as 4; and “Strongly agree” or “Totally agree” as 5. For categorical variables, one-hot encoding was applied to nominal variables (e.g., Gender) and ordinal encoding for ordinal variables (e.g., Education). No normalization or standardization was performed for tree-based models, such as Random Forest and XGBoost. However, this step was necessary for scale-sensitive models, such as Logistic Regression. The models were validated to ensure robustness and avoid overfitting. For the train–test division, the data were divided into training and test sets using a stratified strategy (80% for training and 20% for testing) to maintain the proportion of the target variable in both sets. K-fold cross-validation (ideally k = 5 or 10) was then used to obtain more stable estimates of the models’ performance and avoid overfitting.

3. Results and Discussion

The survey consisted of 39 questions related to sociodemographic information (questions 1 to 8), environmental conditions (questions 9 to 15), infrastructure and safety (questions 16 to 22), personal perception and attitudinal barriers to bicycle use (23–27), and practical and logistical barriers to bicycle use (questions 28 to 39).

3.1. Sociodemographic Information

A total of 2077 responses were recorded. Of these, 2068 participants (99.6%) agreed to take the survey, while only nine (0.4%) refused. The data collection period lasted 62 days, and the average response time was 15 minutes.

Of the 2068 total respondents, 52% (1065 participants) identified as female and 47% (982 participants) identified as male. Additionally, 20 participants identified as non-binary, and one identified as “other.” Participants’ ages range from 22 to 78 years old, with an average age of 29.56 years old. The mode age is 22 years old, recorded 217 times among respondents.

The participants’ academic backgrounds revealed that the undergraduate level was the most common, with 880 responses. This was followed by secondary school (403 responses), technical school (250 responses), and technologist (234 responses). Among postgraduate respondents, 189 were enrolled in postgraduate studies, 82 had master’s degrees, and 19 had doctorates. Primary school was the least represented level with only 11 participants.

Regarding population type, most respondents (90%, or 1853) reside in urban areas, while 10% (215) reside in rural areas. Socioeconomic stratification showed that Stratum 3 had the most participants (770), followed by Socioeconomic Level 2 (671) and Socioeconomic Level 1 (371). Socioeconomic Levels 4, 5, and 6 had 186, 53, and 17 participants, respectively.

3.2. Environmental Conditions

To evaluate this aspect, questions 9 to 15 of the applied survey were analyzed (Figure 2):

9.

The places are too far away to cycle to.

10.

My journey is too short to consider cycling.

11.

It would take too long to travel from my place of residence to my final destination.

12.

The slopes are too steep.

13.

The weather conditions are not conducive to cycling.

14.

There are many bad smells and exhaust fumes on the road.

15.

Cycling is good for the environment.

Questions 9, 10, 11, 12, 13, and 14 explore various perceived barriers to bicycle use. First, the data show that distance and perceived travel time are significant barriers. A considerable proportion of respondents (above 68%) believe that “places are too far away to cycle to.” Nearly 62% of respondents consider their trips too short to consider cycling, and nearly 62% believe it would take too long to travel from their place of residence to their final destination. These seemingly contradictory responses are not uncommon in mobility studies [25,26]. They suggest that distances can be an impediment because they are "too far" and a reason not to consider cycling if the trip is “too short” to justify the effort or if the estimated travel time is perceived as excessive, regardless of the actual distance. This highlights the importance of subjective perception of distance and time [27].

In addition, topographical and environmental characteristics emerge as deterrent factors. Above 61% of respondents agreed that “the slopes are too steep,” suggesting that the terrain poses a significant physical challenge to potential users. “Weather conditions” are also an important factor, with above 63% agreeing or strongly agreeing that they hinder bicycle mobility. Additionally, more than 62% of respondents perceive that “there are many bad odors and exhaust fumes on the road.” Together, these elements create an environment that is perceived as inhospitable to cycling [28].

Despite these operational and contextual barriers, there is a strong consensus on the environmental benefits of cycling. Above 91% of respondents agree or strongly agree that “cycling is good for the environment.” This finding is significant because it shows that although there are practical and perceived comfort and safety barriers, awareness of cycling’s positive impact is high.

3.3. Infrastructure and Safety

The design of transportation infrastructure, particularly bicycle infrastructure, directly influences safety and people’s decisions to use bicycles as a mode of transportation. The type of infrastructure can guarantee greater or lesser safety, which encourages or discourages bicycle use in urban and non-urban areas. In developed countries such as the United States [29], Canada [30], Italy [31], and the Netherlands [32], cyclists prefer environments that physically separate them from vehicular traffic, such as bike lanes with vertical barriers. China, a country with a long history of bicycle use, shares these preferences and desires cleaner air [33]. In most cases, how safe a place feels is directly related to how much interaction there is with motor vehicles [34].

To evaluate this aspect, questions 16 to 22 of the applied survey were analyzed (Figure 3):

16.

There are no bike lanes, and the ones that exist are of very poor quality.

17.

There are too many intersections, and the intersections are not safe for cyclists.

18.

The roads are too narrow for cycling to be safe.

19.

Street lighting is poor.

20.

Bicycle parking facilities are inadequate.

21.

There is too much traffic, and traffic moves too fast for cycling to be safe.

22.

Drivers do not respect cyclists.

Analyzing the perception of infrastructure and road behavior (questions 16–22) is essential for identifying critical barriers and areas for intervention [29]. Item 16 reveals that inadequate cycling infrastructure is a significant obstacle to bicycle use. The high percentages of respondents who answered “Strongly agree” or “Agree” suggest that the absence or poor quality of bike lanes creates a perception of insecurity and discourages bicycle use. This finding aligns with the literature on active mobility, which emphasizes that segregated infrastructure is a key factor in enhancing the perception of safety and promoting bicycle use [29,35]. Item 17 reveals that road crossings are critical points of perceived risk. The concentration of responses in the “agree” category underscores the need for safer and clearer intersection designs for cyclists, such as specific traffic lights, adequate horizontal and vertical signage, and roundabouts with cyclist priority. According to a study conducted by McNeil et al. and Campisi et al., these measures have the potential to reduce stress levels among cyclists [29,31]. The perception of narrow roads is a relevant factor, given that, in many Colombian cities, cyclists must share lanes with motorized vehicles (there are no dedicated lanes), or the bike lane is poorly marked. This exacerbates the fear of interacting with motorized vehicles. These findings underscore the necessity of implementing speed reduction policies, traffic calming measures, and, ideally, reallocating road space to create protected cycling infrastructure; according to McNeil et al., this would encourage potential or cautious cyclists to use bicycles as a means of transportation [29]. In addition to the above, the responses to item 21 show that speed and traffic volume directly intimidate cyclists. These findings validate the need for traffic management strategies that include reduced speed zones (30 km/h), traffic calming measures, and driver awareness campaigns about the vulnerability of cyclists [36,37]. One factor that exacerbates the perception of road safety is the lack of respect that drivers have for cyclists (item 22). This factor is possibly the most critical and multifactorial barrier. The predominance of responses indicating “strong agreement” reflects poor road culture and the urgent need for intensive road safety education programs and awareness campaigns that promote coexistence and mutual respect among all road users. There is also an urgent need for stricter enforcement of traffic rules to protect vulnerable users.

Two relevant infrastructure factors are public lighting and bicycle parking facilities. Item 19 indicates that inadequate lighting negatively affects the perception of road and personal safety. It is a significant deterrent to cycling in low visibility conditions. Therefore, investing in quality public lighting on cycle routes is essential. The second factor, bicycle parking facilities (item 20), is a significant yet often overlooked barrier. Cyclists need to be confident that their bicycles will be safe at their destinations. This requires installing modern, visible, secure, well-located, and preferably protected or monitored bicycle parking facilities [28].

3.4. Personal Perception and Attitudinal Barriers to Bicycle Use

To evaluate this aspect, questions 23 to 27 of the applied survey were analyzed (Figure 4):

23.

I don’t have access to a bicycle.

24.

I have too many bags to carry/my bags are too heavy.

25.

It requires too much advance planning.

26.

I wouldn’t be able to fix minor mechanical problems (e.g., repairing a flat tire or adjusting the brakes).

27.

Traveling by other means of transportation is more comfortable.

First, the lack of direct access to a bicycle is a fundamental barrier. Many people do not own a bicycle, which limits its potential use. This underscores the need for affordable access programs, such as robust bike-sharing systems or purchase subsidies.

In terms of practicality, the need to transport items is a clear impediment. Cyclists often require cargo solutions, such as panniers, baskets, or trailers, but these are not always accessible or convenient. This underscores the importance of designing bicycles and cargo transport solutions that adapt to users’ everyday needs [38].

The perception that cycling “requires too much advance planning” (item 25) suggests a lack of spontaneity and flexibility, which could deter potential users. This perception may be related to the need to plan safe routes, consider appropriate clothing, and anticipate bicycle management at the destination [39].

A critical aspect of a cyclist’s autonomy and empowerment is the ability to solve minor mechanical problems (item 26). The high level of agreement on this issue suggests a lack of basic maintenance skills. Without proper training, this can lead to significant insecurity and dependence, which can discourage longer or solo trips. Providing education in basic mechanics and making public repair stations available could be effective solutions.

Finally, item 27 states that “traveling by other means of transportation is more comfortable” and is a synthesis of many perceived barriers. It groups together factors such as exposure to the weather, physical effort, luggage management, and the perception of lower safety and convenience compared to motorized or public transportation. Unless the barriers that undermine cycling’s comfort and convenience are addressed, it will always be at a disadvantage compared to other modes of transport that offer greater perceived comfort [40].

3.5. Practical and Logistical Barriers to Bicycle Use

This section of the survey reveals several personal perception and significant attitudinal barriers that discourage cycling. These subjective factors are key to understanding why some people choose not to use bicycles, even when infrastructure improves.

To evaluate this aspect, questions 28 to 39 of the applied survey were analyzed (Figure 5):

28.

It is not convenient due to my other activities.

29.

I am not fit enough to ride a bike.

30.

I don’t feel safe riding a bike.

31.

I would get very hot and sweat a lot if I rode a bike.

32.

I am often too tired to ride a bike.

33.

My clothes are not suitable for riding a bike.

34.

Cycling would ruin my hair, especially if I wore a helmet.

35.

I would be afraid of being attacked by harassers or strangers along the way.

36.

It is not appropriate to ride a bike to my usual destinations.

37.

I am too lazy to ride a bike.

38.

Walking is more sociable.

39

Driving or being driven in a car is more fun.

A significant number of these barriers relate to the convenience of integrating cycling into daily life. A high proportion of respondents agree with the statement, “It is not convenient due to my subsequent activities” (item 28), which underscores the need for intermodal solutions and supporting infrastructure at destinations, such as showers, changing rooms, and secure parking, to facilitate the transition between cycling and other activities. This perception is exacerbated by concerns about physical comfort and personal image. Items 31, 33, and 34 reflect these concerns: “I would get very hot and perspire/sweat a lot if I rode a bicycle,” “My clothes are not suitable for cycling,” and “Cycling would ruin my hair, especially if I wore a helmet.” These concerns can be mitigated by promoting cycling as part of an active and healthy lifestyle and by providing better end-of-trip facilities [41].

In terms of perceived physical ability and effort, items 29 and 32, which ask, “I am not fit enough to ride a bike” and “I often feel too tired to ride a bike,” respectively, show that subjective physical condition is a barrier for many. Therefore, bicycle promotion initiatives should focus on infrastructure as well as programs that gradually encourage physical conditioning, highlight health benefits, and promote accessible routes for all skill levels. Additionally, “laziness” (item 37: “I am too lazy to ride a bike”) is an attitudinal factor that competes with the effort of cycling [42].

The item “I don’t feel safe riding a bicycle” (item 30) reflects personal safety and perceived risk, and its high incidence is concerning. Similarly, the fear of being attacked by harassers or strangers while riding is a critical barrier, especially for certain demographic groups. Improvements in road infrastructure, such as segregated bike lanes and safe crossings, are necessary, as are a greater security presence, improved urban lighting, and the promotion of routes with greater social visibility.

Finally, subjective preferences and the appeal of other modes of transportation act as direct competitors. The perception that “it is not appropriate to cycle to my regular destinations” (item 36) suggests social or normative barriers. Item 38, “Walking is more sociable,” highlights that some people value social interaction above the efficiency of cycling. Item 39, “Driving a car or being driven is more fun,” shows that the comfort and enjoyment of motorized transportation remain strong attractions [40].

3.6. Evaluating Predictive Models

The three predictive models below estimate the probability that an individual is a potential bicycle user based on demographic variables and perceived barriers to use.

Table 3. Comparative evaluation of model performance using cross-validation (Logistic Regression, Random Forest, and XGBoost).

Model	Logistic Regression	Random Forest	XGBoost
Accuracy	0.91032 ± 0.01259	0.95202 ± 0.00871	0.99224 ± 0.00283
Precision	0.86733 ± 0.02831	0.97324 ± 0.01651	0.99105 ± 0.01194
Recall	0.84955 ± 0.01878	0.87393 ± 0.02538	0.98481 ± 0.01071
F1-Score	0.85811 ± 0.01918	0.92062 ± 0.0149	0.98781 ± 0.00441
ROC–AUC *	0.97206 ± 0.00651	0.99388 ± 0.00175	0.99983 ± 0.00014

* Area Under the Receiver Operating Characteristic Curve.

shows the performance evaluation of the three models used in the analysis.

Examining the performance results reveals a clear hierarchy in the effectiveness of the implemented predictive models. While all three models perform well, XGBoost stands out as the best analysis tool for this dataset due to its superior performance. Logistic Regression has the most modest metrics, with an accuracy of 0.91032 ± 0.01259. This indicates that it correctly classifies 91% of individuals [43]. Its precision of 0.86733 ± 0.02831 suggests that 87% of the individuals it predicted as “potential users” were actually users. Its recall (sensitivity) of 0.84955 ± 0.01878 indicates that it correctly identified 85% of all true “potential users” in the sample [44]. The F1-score, which balances precision and recall, is 0.85811±0.01918 [45]. Its ROC–AUC of 0.97206 ± 0.00651 demonstrates its ability to distinguish between the two classes is very good [46].

The Random Forest model marks a significant improvement in performance. Its accuracy increases to 0.95202 ± 0.00871, and its precision rises to 0.97324 ± 0.01651. This means its predictions of “potential users” are highly reliable. Although its recall is slightly higher than logistic regression’s (0.87393 ± 0.02538), the significant increase in precision raises its F1-score to 0.92062 ± 0.0149, indicating a good balance. With an ROC–AUC of 0.99388 ± 0.00175, the model nearly achieves perfect discrimination between users and non-users.

The XGBoost model finally exhibits the best performance. With an accuracy of 0.99224 ± 0.00283, precision of 0.99105 ± 0.01194, recall of 0.98481 ± 0.01071, and F1-score of 0.98781 ± 0.00441, it is a robust model. A precision score of 1.0 means that there were no false positives; every person identified as a “potential user” by the model was actually one. Its recall indicates that it barely missed any true potential users. The maximum possible ROC–AUC of 0.99983 ± 0.00014 confirms its discriminatory power for this analysis.

Overall, the consistency and robustness of the XGBoost model, followed closely by Random Forest, confirm that the patterns identified in the data are not random. These advanced models can capture complex nonlinear interactions between variables and provide sufficient confidence that the factors they identify as important are the most decisive predictors of the decision to use a bicycle.

According to the results of the Logistic Regression model (

Table 4. Results of the logistic regression model: coefficients and odds ratios of barriers to bicycle use.

Variable	Coefficient ($\mathit{\beta }$)	Odds Ratio
Academic Training_Undergraduate	0.2995	1.3492
The slopes are too steep	0.2319	1.2610
My trip is too short to consider biking	0.2154	1.2403
I am too lazy to ride a bike	0.2065	1.2294
Academic Training_Technologist	0.2059	1.2286
Places are too far to go by bike	0.2018	1.2236
The public lighting is poor	0.1875	1.2062
Population Type_Urban	0.1791	1.1961
There are no bike lanes or they are of very poor quality	0.1738	1.1898
Cycling would ruin my hair, especially if I wore a helmet	0.1446	1.1555
It involves too much advance planning	0.1349	1.1444
Academic Training_Postgraduate	0.1141	1.1208
There is too much traffic on the roads/traffic is too fast for biking to be safe	0.1115	1.1180
Academic Training_Secondary	0.1105	1.1168
There are many bad smells and exhaust fumes on the road	0.0931	1.0975
Gender: Female	0.0854	1.0891
I have to carry too many bags/my bags are too heavy	0.0695	1.0719
Gender: Male	0.0492	1.0504
Academic Training_Technician	0.0471	1.0483
I often feel too tired to ride a bike	0.0427	1.0437
Bicycle parking facilities are not good	0.0298	1.0302
Riding a bike is good for the environment	0.0038	1.0038
Traveling by other means of transport is more comfortable	−0.0593	0.9424
Socioeconomic Stratum	−0.0793	0.9238
Academic Training_Primary	−0.1023	0.9028
The roads are too narrow for biking to be safe	−0.1037	0.9015
Driving a car or being driven is more fun	−0.1119	0.8942
Biking to my usual places is not appropriate	−0.1486	0.8619
Age	−0.1556	0.8559
Walking is more sociable	−0.1590	0.8530
My clothes are not suitable for biking	−0.1670	0.8462
The trip from my residence to the final destination would take too long	−0.2035	0.8159
I could not fix minor mechanical problems (e.g., repair a flat tire or adjust the brakes)	−0.2075	0.8126
I would be afraid of being attacked by harassers or strangers on my way	−0.2225	0.8005
Academic Training_Master’s Degree	−0.2636	0.7683
Weather conditions do not favor bicycle mobility	−0.265	0.7672
There are too many crossings/The crossings are not very safe for cyclists	−0.3266	0.7214
I am not fit enough to ride a bike	−1.2144	0.2969
Drivers do not respect cyclists	−1.3012	0.2722
I do not have access to a bicycle	−1.3144	0.2686
I would get very hot and sweat a lot if I rode a bike	−1.6971	0.1832
I do not feel safe riding a bike	−1.8574	0.1561
It is not convenient because of my subsequent activities	−1.9566	0.1413

), the most significant predictors of being classified as a potential bicycle user are strongly associated with convenience and safety. The three most powerful predictors that deter potential users have nothing to do with the trip itself but rather its consequences. The predictor “It is not convenient because of my subsequent activities” (odds ratio = 0.1413) is the most devastating, reducing the likelihood of being a potential user by 86%. It reduces the likelihood of becoming a user by 86%. From a behavioral perspective, this represents “logistical friction” [47]. The person may be evaluating not just the A–B journey but their entire daily itinerary. Therefore, a possible lack of end-of-journey facilities (showers, changing rooms, and secure parking) can transform a transportation solution into a logistical problem. The second predictor is “I would get very hot and sweat a lot if I rode a bike” (odds ratio = 0.1832). This is the physiological and component of “social friction” [48]. Sweating is not only physically uncomfortable; it is also a social barrier. In many work and social cultures, arriving sweaty or disheveled is considered unprofessional or inappropriate. Combined with the lack of convenience, this factor can create an psychological barrier. The third predictor is “I don’t feel safe riding a bike” (odds ratio = 0.1561). This is the “emotional friction” [49]. Feeling vulnerable may create constant cognitive stress that negates the benefits of cycling, such as exercise and savings. Interestingly, certain variables indicating poor infrastructure have an odds ratio (OR) greater than one, suggesting they increase the likelihood of being a potential user. Examples include “The slopes are very steep” (OR = 1.2610), “Street lighting is poor” (OR = 1.2062), and “There are no bike lanes or they are of very poor quality” (OR = 1.1898). One possible explanation is related to cognitive biases [50]. Complaining about infrastructure indicates interest because a person who would never use a bicycle is indifferent to the quality of bike lanes or lighting. In contrast, a potential bicyclist may criticize the environment. Perceiving poor infrastructure quality may indicate that they have evaluated the routes and identified problems. They are “potential” precisely because they are already thinking like cyclists. Thus, safety-conscious cyclists may simply avoid riding at night if the lighting is poor. A potential user has already developed strategies to mitigate these risks, such as avoiding certain hills or only riding during the day. Therefore, although they recognize the problem, they do not consider it an absolute impediment that would classify them as “non-potential.”

Other significant barriers include “I don’t have access to a bicycle” (OR = 0.2686) and “I’m not physically fit enough” (OR = 0.2969). These results highlight the need for policies that address access (e.g., public bicycle systems and subsidies) and self-efficacy (e.g., training programs and beginner-friendly routes). Perception of one’s own ability is a determining factor in behavior.

The analysis of the importance of variables in the Random Forest model (

Table 5. Importance of predictor variables according to the Random Forest model.

Variable	Importance
It’s not convenient due to my later activities	0.1429
I don’t feel safe riding a bicycle	0.1353
I would get too hot and sweat a lot if I rode a bicycle	0.0946
I’m not fit enough to ride a bicycle	0.0734
I often feel too tired to ride a bicycle	0.0639
I don’t have access to a bicycle	0.0410
Traveling by other means of transport is more comfortable	0.0324
My clothes are not suitable for cycling	0.0316
Drivers do not respect cyclists	0.0270
Driving a car or being driven is more fun	0.0263
Cycling would ruin my hair, especially if I wore a helmet	0.0260
It’s not okay to bike to my regular places	0.0222
I would be afraid of being attacked by harassers or strangers on my way	0.0211
Age	0.0180
I am too lazy to ride a bicycle	0.0171
I couldn’t fix minor mechanical problems (e.g., fix a flat tire or adjust the brakes)	0.0170
I have to carry too many bags/my bags are too heavy	0.0158
Weather conditions are not favorable for cycling	0.0152
The hills are too steep	0.0141
It involves too much planning ahead	0.0135
The roads are too narrow for cycling to be safe	0.0118
Walking is more sociable	0.0113
There is too much traffic on the roads/traffic is too fast for cycling to be safe	0.0100
There are too many intersections/The intersections are not very safe for cyclists	0.0099
The trip from my place of residence to the final destination would take too long	0.0093
Bicycle parking facilities are not good	0.0087
Socioeconomic stratification	0.0085
The street lighting is poor	0.0083
There are no bike lanes or they are of very poor quality	0.0083
The places are too far to go by bike	0.0079
There are many bad smells and exhaust fumes on the road	0.0067
Riding a bike is good for the environment	0.0065
My journey is too short to consider cycling	0.0063
Department (Currently lives in)_Huila	0.0040
Department (Currently lives in)_Magdalena	0.0039
Academic background_Undergraduate	0.0037
Gender_Male	0.0035
Department (Currently lives in)_Cundinamarca	0.0029
Academic background_Technician	0.0027
Academic background_Technologist	0.0023
Academic background_High School	0.0022
Population type_Urban	0.0020
Academic background_Postgraduate	0.0016
Academic background_Master’s Degree	0.0010
Gender_Female	0.0005

) quantifies the predictive power of each factor to determine whether an individual is a potential bicycle user [51]. The results indicate that personal and convenience barriers are the most influential predictors, surpassing infrastructure factors. The most important variable is “It is not convenient due to my subsequent activities” (0.1429), followed by “I do not feel safe riding a bicycle” (0.1353) and “I would get very hot and sweat a lot” (0.0946). This shows that the decision to use a bicycle is strongly influenced by how easily the trip can be integrated into the daily routine and by how safe the individual perceives it to be. Self-assessed physical condition (“I am not fit enough,” 0.0734) and fatigue (“I often feel too tired,” 0.0639) are also highly predictive, revealing that physical condition and energy levels are key determinants. Notably, structural barriers, such as a lack of access to a bicycle and the perception that drivers do not respect cyclists, also rank high, albeit with a lower weighting than personal convenience barriers. This suggests that although infrastructure and road culture remain relevant, the most effective strategies to promote cycling should primarily target personal and logistical challenges.

The analysis of variable importance using the XGBoost model (

Table 6. Importance of predictor variables according to the XGBoost model.

Variable	Importance
I don’t feel safe riding a bicycle	0.2942
It’s not convenient due to my later activities	0.1760
I would get too hot and sweat a lot if I rode a bicycle	0.1401
I’m not fit enough to ride a bicycle	0.0746
I don’t have access to a bicycle	0.0491
Drivers do not respect cyclists	0.0428
I often feel too tired to ride a bicycle	0.0399
Cycling would ruin my hair, especially if I wore a helmet	0.0202
I couldn’t fix minor mechanical problems (e.g., fix a flat tire or adjust the brakes)	0.0157
Driving a car or being driven is more fun	0.0128
I am too lazy to ride a bicycle	0.0119
My clothes are not suitable for cycling	0.0114
Weather conditions are not favorable for cycling	0.0103
The places are too far to go by bike	0.0095
It involves too much planning ahead	0.0093
My journey is too short to consider cycling	0.0092
Bicycle parking facilities are not good	0.0085
Age	0.0077
Academic background_Undergraduate	0.0077
The trip from my place of residence to the final destination would take too long	0.0076
Riding a bike is good for the environment	0.0075
The roads are too narrow for cycling to be safe	0.0060
Walking is more sociable	0.0051
Traveling by other means of transport is more comfortable	0.0048
Department (Currently lives in)_Atlántico	0.0044
There are many bad smells and exhaust fumes on the road	0.0041
There are no bike lanes or they are of very poor quality	0.0033
Socioeconomic stratification	0.0027
Academic background_Postgraduate	0.0019
The hills are too steep	0.0018
The street lighting is poor	0.0000
There are too many intersections/The intersections are not very safe for cyclists	0.0000
I have to carry too many bags/My bags are too heavy	0.0000
It’s not okay to bike to my regular places	0.0000
I would be afraid of being attacked by harassers or strangers on my way	0.0000
Gender_Male	0.0000
Academic background_Master’s Degree	0.0000
Gender_Female	0.0000
Academic background_Primary School	0.0000
There is too much traffic on the roads/Traffic is too fast for cycling to be safe	0.0000
Academic background_High School	0.0000
Academic background_Technologist	0.0000
Population type_Urban	0.0000
Academic background_Technician	0.0000

) confirms and reinforces the findings of other models [52]. As detailed in Table 3, the analysis highlights subjective and convenience barriers above infrastructure factors, which have great predictive power. The XGBoost model, which uses an advanced gradient boosting framework, unequivocally identifies the perception of safety as the most decisive factor. “I don’t feel safe riding a bicycle” is the most important factor, with a value of 0.294. Logistical inconvenience (“It is not convenient due to my subsequent activities”) and concern for physical comfort (“I would get very hot and sweat a lot”) follow in order of relevance with values of 0.176 and 0.140, respectively. These three factors alone account for most of the model’s predictive weight, highlighting that the decision to use a bicycle is fundamentally governed by perceived risk and ease of integrating the trip into everyday life. Self-perceived physical fitness (“I’m not fit enough,” 0.075) and lack of access to a bicycle (0.049) are also highly significant variables. Notably, many infrastructure-related variables, such as the quality of bike lanes and street lighting, are given no importance in this model. This does not mean that these variables are unimportant but rather that their ability to predict potential users is overshadowed by the aforementioned personal barriers. From a public policy perspective, this suggests that, to expand the cycling population, interventions must prioritize mitigating the perception of insecurity and resolving logistical and comfort issues related to travel.

The results of the Logistic Regression, Random Forest, and XGBoost models (Table 4, Table 5, and Table 6, respectively) show that personal, convenience, and safety perception barriers are more relevant than infrastructure factors when evaluating bicycles as a means of transportation. Variables such as “It is not convenient due to my subsequent activities,” “I do not feel safe riding a bicycle,” and “I would get very hot and sweat a lot” consistently rank at the top in all three models. This holds true whether they are measured by their low odds ratio (logistic regression) or their high importance score (Random Forest and XGBoost). This consistency across models underscores that these three elements constitute the most powerful core barriers. Additionally, factors such as self-perceived unfitness, lack of access to a bicycle, and interaction with traffic (“Drivers don’t respect cyclists”) repeatedly appear in critical positions in all analyses, forming a second level of barriers. Equally significant is the fact that, in machine learning models (Random Forest and XGBoost), variables related to physical infrastructure, such as the quality of bike lanes, street lighting, and road width, have considerably less predictive importance. This cross-validation of different analytical techniques provides compelling evidence that interventions must focus primarily on mitigating subjective and logistical barriers to convert potential users into active cyclists. This demonstrates the need for a public policy approach that goes beyond simply providing infrastructure.

4. Conclusions

The research demonstrated strong convergence in the results of the artificial intelligence predictive models (Logistic Regression, Random Forest, and XGBoost). These results revealed that the most significant barriers to bicycle use are not primarily infrastructure-related but rather personal, logistical, and related to perceptions of safety. These findings highlight the inadequacy of public policies focused exclusively on constructing bike lanes to convert potential users into active cyclists. The three predictors that consistently ranked as the most relevant were as follows: (a) Inconvenience due to subsequent activities: “It’s not convenient due to my subsequent activities.” With an Odds Ratio of 0.1413, this factor is the most decisive, reducing the probability of being a potential user by 86%. This suggests that a lack of support infrastructure at destinations, such as showers and secure parking, creates logistical friction that negates the benefits of cycling. (b) Perception of insecurity: “I don’t feel safe riding a bicycle.” This predictor was associated with high emotional friction because the feeling of vulnerability significantly reduces the likelihood that someone will become a user (odds ratio of 0.1561). In the XGBoost model, it was the factor with the greatest predictive power, at 0.2942. (c) Physical discomfort: “I would get too hot and sweat a lot.” This physiological and social barrier reduces the likelihood of use by 82% (odds ratio of 0.1832). This emphasizes the importance of addressing personal comfort factors that conflict with the desire for a professional or socially acceptable image. These results underscore the necessity of a more comprehensive approach to designing mobility policies. Instead of focusing solely on infrastructure, interventions should prioritize eliminating subjective and logistical barriers that hinder bicycle use. Comprehensive, people-centered strategies are necessary to improve travel feasibility, perceived safety, and user comfort.