Wearable Internet of Things Gait Sensors for Quantitative Assessment of Myers–Briggs Type Indicator Personality

Gait is a typical habitual human behavior and manifestation of personality. The unique properties of individual gaits may offer important clues in the assessment of personality. However, assessing personality accurately through quantitative gait analysis remains a daunting challenge. Herein, targeting young individuals, standardized gait data are obtained from 114 subjects with a wearable gait sensor, and the Myers–Briggs Type Indicator (MBTl) personality scale is used to assess their corresponding personality types. Artificial intelligence algorithms are used to systematically mine the relationship between gaits and 16 personality types. The work shows that gait parameters can indicate the personality of a subject from the four MBTI dimensions of E‐l, S‐N, T‐F, and J‐P with a concordance rate as high as 95%, 96%, 91%, and 91%, respectively. The overall measurement accuracy for the 16 personality types is 88.16%. Moreover, a personality tracking experiment on all the subjects after one year to assess the stability of their personality is also conducted. This research, which is based on a smart wearable Internet of Things gait sensor, not only establishes a new connection between behavioral analysis and personality assessment but also provides a set of accurate research tools for the quantitative assessment of personality.

advanced sensing and AI-related technologies to develop a systematic and accurate measurement tool for assessing personality characteristics.
People's behaviors are closely related to their personality traits. [1,9]Behavioral analysis has powerful practical applications in mental health therapy and organizational psychology.The use of habitual behavioral patterns to assess personality can effectively prevent the interference of the social desirability effect and other subjective factors. [10]Besides, behavior can be observed, recorded, and analyzed systematically with the aid of technological devices.[13][14][15][16] For instance, research teams from Nagoya University and Tsinghua University examined the relationship between facial behavior and personality. [11,12,15]Other research teams, such as those from MIT, investigated the relationship between voice and personality. [13,14,16]However, owing to various confounding factors, including gender and age effects, the concordance rate of such physiological indicators, with conventional psychometric personality measures, is low.Hence, systematically quantifying and evaluating the various dimensions of personality based on behavioral information from faces and voices may be difficult, and other behavioral indicators of personality should be identified to increase accuracy.
Gait is one of the salient features of human behavior, which is generated by the combined action of the brain and nerves. [17]Gait also reflects an individual's cognition, character, etc., and provides clues to individuals' mental and health conditions. [18]ompared with other biometrics, gait is difficult to camouflage; thus, it can offer a highly objective measurement.Collado-Vázquez et al. believed that gait reflects internal characteristics, [19] while others showed that walking speed in adulthood can reflect personality to a certain extent. [20]For example, two dimensions of the Big Five Personality Traits are generally related to gait speed and reduced gait speed. [21]Specifically, amplitude of upper and lower extremity movements and walking speed are associated with aggression. [22]However, the analysis of gait and personality in the aforementioned studies was not systematic and comprehensive, and gait was mostly limited to walking velocity.
[25][26][27] Examples of recent work using motion and pressure sensor data to relate gait features to human characteristics are shown in Table 1.Experiments at the Newcastle Neuroscience Institute demonstrated that gait is highly reliable in judging personality traits. [28]A research team from Shanghai Jiao Tong University used the Kinect system to discover gait characteristics that may be related to personality. [29]Another team from Changwon National University in South Korea used specific electric charges (GaitRite program) to understand the gait patterns of the Myers-Briggs Type Indicator (MBTI) personality types. [30]Meanwhile, a psychology team at Carleton University applied gait research to crime detection analysis. [31]Most recent studies examined the correlation between gait and personality, but no systematic and quantitative evaluation on how gait can explain personality preferences exists.Hence, a comprehensive examination of the processes behind gait-based patterns is necessary to advance our understanding of personality to a considerable extent.
Currently, there are many assessment tools to describe personality and the most prominent and influential methods are 1) Cattell's 16-factor personality model, 2) the three dimensions of Eysenck Personality (Eysenck Big Three) model, 3) the MBTI, and 4) the Big Five assessment.Perhaps the most important reason that Cattell's 16-factor model never gained full academic acceptance is that it is harder to understand than simpler models such as the Eysenck Big Three or the Big Five; however, Eysenck's model is insufficient to account for the complexity of the wide range of human personalities. [32]Currently, the Big Five factors and the MBTI are the two most commonly used personality models, both of which are based on Carl Jung's inside-out dichotomy and provide personality insights.In essence, the Big Five assessment measures how many traits a person has and is a feature-based approach, [33] while the MBTI assesses what preferences each person has on dominant functions and represents a choice of how people think and behave-which is a type-based approach that is consistent with the way humans choose to walk.
The MBTI is a psychometric inventory for assessing individuals' preferences in the dominant function, representing their choice in the way they think and behave and exhibiting consistency with the way they choose to walk.The MBTI is a typebased approach first developed by American psychologist Katherine Cook Briggs (1875-1968) and her psychologist daughter, Isabel Briggs Myers, based on the theory of mental types of the renowned psychoanalyst Carl G. Jung and their long-term observations and research on differences in human personality. [34,35]The MBTI serves as a reliable aid for defining personality types based on adequate research and validation, [36] including applications in career development and team building. [37,38]tructurally, the division of the MBTI into four dimensions is a natural classification method, which matches machine learning technology and is naturally friendly in terms of technical application.Owing to its practicality, the MBTI is widely used in industries for personnel selection, career planning, and talent development.This evaluation model is used in 115 countries and available in 29 languages and demonstrated satisfactory reliability and validity in recent years. [39,40]Figure 1 presents the MBTI 16 personality composition and structural chart.In the innermost circle lies the four worldviews, each with two dimensions.The outer circles include the 16 personality types around four pairs of human worldview categories. [41]Myers described the two dimensions of the four worldviews as "preference" in four pairs of categories: extraverted (E)-introverted (I), sensory (S)-intuitive (N), thinking (T)-feeling (F), and judging (J)-perceiving (P).The differences in the characteristics of the four dimensions are not mutually exclusive but habitual preferences, which also correspond well to Holland's "thing-person" dichotomy.Each of the 16 personality types is associated with a specific pattern of personality traits.
2,43] However, from the experimental point of view, it is mostly limited to high-tech gait laboratories or expensive complex systems.With the growing demand for portable and precise analysis systems, [44] wearable technology has transformed the accessibility of gait analysis, providing an opportunity to assess human behaviors outside the laboratory. [45,46]hus, such technology is becoming increasingly popular and advancing toward mainstream development.In human activity recognition, standard datasets that can be used for learning can be customized. [47]herefore, in this study, we develop a single microelectromechanical systems (MEMS) inertial sensor (with wireless real-time transmission, small and portable, low cost, and fast response) that can support the parallel acquisition of the motion states of multiple objects and be used for standard data acquisition of gait.As shown in Figure 2, the physical features and capability of our proposed system are compared with other previous studies using sensors.The figure clearly indicates the advantages of our proposed system in terms of weight, volume, number of sensors, and "trait detection" functionality over existing systems.
In this study, we demonstrated a one-to-one correspondence between collected standardized gait data and personality type using only IMU sensors.By quantifying gait kinematic parameters to analyze the gait performance of groups of different or the same personality types, we systematically describe the different preferences of the MBTI 16 personality types.The four preference dichotomies include the source and focus of energy (extroverted-introverted), the way of understanding the world and information (sensory-intuitive), the benchmark for judging the world (thinking-feeling), and the way of processing or coping with the external world (judging-perceiving).These four aspects provide theoretical explanations for clarifying the relationship between gait and personality.Next, we extracted gait characteristics from IMU data as predictive factors and achieved an accurate description of personality preferences and measurement of 16 personality types through machine learning algorithms.This significantly enhanced our understanding of personality.We successfully conducted research on human personality using IMU data and proposed a method for modeling IMU data based on gait and personality using machine learning.

Experimental Section 2.1. Materials
The framework of the system approach for assessing personality based on the self-developed intelligent wearable Internet of Things (IoT) sensor is shown in Figure 3.The first box illustrates the data acquisition platform (part A of Figure 3), the second box presents the data processing and analysis (part B), and the last box is the machine learning algorithmic prediction (part C).
We obtained the data for this study from 114 university students.Specifically, we collected gait data from all the participants who completed an online questionnaire containing the MBTI and personal information.This study was approved by the Research Ethics Committee (Northeastern University Ethics Committee), and all the participants indicated informed consent.We used several exclusion criteria for collecting the data, specifically, we limited the age of the participants to 18-28 years and set the education level to cover all stages, from the undergraduate level to the postgraduate level, and the health requirement to having no history of nerve, muscle, bone, and other diseases.A summary of the subjects' age, height, weight, and body mass index is provided in Table 2.
In addition to the individual circumstances, to properly capture the reasonable relationship between walking and personality, considering the actual walking environment is crucial, including the weather, weight, route, obstacles, and so on.As the gait of the subjects differed under the observation of different numbers of experimenters, [48] we designed a 50 m Â 26 m rectangular route in the school and MEMS-based sensor, as shown in Figure 4, and informed the subjects to not bear weight.The sensor was worn on the right ankle to achieve real-time normalized gait data acquisition.The sensor is composed mainly of a six-degree-of-freedom micro-IMU and new low-power microcontroller N52832 that can support Bluetooth.The IMU employs the MPU9250 chip, which integrates a 3-axis gyroscope and a 3-axis  accelerometer.The entire sensor is a 12.0 mm Â 12.5 mm Â 1 mm miniature device, its power consumption is as low as 5 mA, its value is accurate to eight decimal places, and it is stable and reliable.We also used the sensor to collect the acceleration and angular rate data of the X-, Y-, and Z-axes.The collected data are transmitted wirelessly to the mobile phone through Bluetooth for storage.After familiarization with the scene, the subjects were asked to walk naturally and independently.At the same time, our acquisition system has a vision sensor, which can transmit the walking picture of the foot to the mobile phone in real-time for saving.The video data are provided for subsequent reference and analysis to verify the validity of the data collected by the IMU sensor.
We divided the personality acquisition process into three stages: pretest explanation, test taking, and posttest communication.
Pretest explanation: One day before the test, we explained the purpose, significance, content scope, type, and specific procedures of the test to the participants.Before the test, we instructed the participants to read the instructions and adjust their emotional and physiological states accordingly.
Test taking: This stage involved the participants taking an online quiz after reading the instructions for each section.
Posttest communication: After the participants completed the test, we synthesized the test results and issued a personality result report to each of the participants (including explanations of the basic dimensions; descriptions of their strengths and possible blind spots, including as partners or parents; career development analysis suggestions; and so on).Finally, we communicated and explained the report results positively and objectively to each participant to ensure maximum understanding.

Data Analysis
The sampling rate of the experimental device used in the experiment was 50 Hz.The main sources of error during the data collection process were the random electronic noise generated by the sensing device and the uncertain jitter phenomenon of the sensor in motion.To reduce the effect of the errors brought about by the noise, we preprocessed the raw data (removed outliers and denoised).In addition, we screened the collected questionnaire information according to indicators such as completion time, degree of completion, and completion speed.The basic information of the sample is shown in Table 2.After the preprocessing, we compared and matched the data to the personality types one by one.Each personality dimension demonstrated differences in the maximum value or time within or between the groups, as shown in Figure 5, which helped in the subsequent data analysis.
We obtained the four-dimension preference index from the questionnaire and divided the participants into 16 personality types.The distribution of the personality types of the subjects is illustrated in Figure 6.

Gait Stability
In exploring the relationship between gait and personality, to ensure the stability of the relationship between the two elements, we conducted tracking experiments on gait and personality.For gait, we used a gait experimental protocol across time, that is, a second gait data acquisition period for all the subjects in the same outdoor environment before and after a year of normal life.We preprocessed the gait data from the second acquisition period and compared them with the data from the previous year.It can be seen from Figure 7 that the gait of the same personality types was relatively stable before and after one year, whereas the gait of different personality types exhibited obvious differences before    and after one year.Nevertheless, some subtle differences were evident in the gait data of the same personality types.

Personality Stability
At the same time, we also conducted a personality tracking experiment on all the subjects after one year in a different environment (i.e., in different classrooms) to assess the stability of their personality.According to the theory of the 16 personality types, stability is positive. [40]In response, we conducted a test-retest reliability assessment on the participants' personality.As shown in Figure 8 and 83% of the subjects (the generally accepted standard in the field is within the range of 70%-90%) demonstrated no personality changes, indicating the positive stability of personality in this study.
During the gait data collection phase, we performed each experiment continuously.To avoid errors during the transition period, we compared the obtained data with the videos recorded during the experiment.We divided the preprocessed time series gait data into two parts, that is, the stance phase and swing phase, and then selected 10 consecutive gait cycles that were relatively stable under natural motion for each individual for the data analysis.

Feature Extraction
First, we extracted the time-domain statistical properties from the preprocessed acceleration and angular acceleration data.Second, we calculated the gait kinematic parameters from the preprocessed acceleration and angular acceleration data.In analyzing the gait parameters, we improved the gait parameter algorithm suitable for the ankle area according to the temporal and spatial variation laws of limb movements when the human body is walking.The algorithm extracted multiple features and described microscopic differences in different types of gait from multiple dimensions.We selected some common gait kinematic parameters (Table 3) including distance parameters such as step length, step length, and foot angle (including pitch angle and roll angle) and time parameters such as single step time, stride time, cadence, pace, stance phase time, swing phase time, gait cycle, ratio of stance phase and swing phase time, and ratio between the two phases and other parameters.We calculated and analyzed the statistical characteristics of some of the parameters.The final dataset included 170 gait parameter features and 20 personality criteria (16 personality types and four preferences) per person.

Model
Based on the gait motion analysis of the above four personality dimensions, we used machine learning algorithms to evaluate  the actual correlation between gait and personality.We took the corresponding gait data as meaningful personality predictors and employed five commonly used machine learning algorithms including decision tree, logistic regression, support vector machine, random forest, and Naive Bayes to predict the personality types.The subset accuracy is 1.0 if the entire predicted label set of a sample strictly matches the true label set, and 0.0 otherwise.If b y i is the predicted value of the ith sample and y i is the corresponding true value, then the proportion of correct predictions divided by the number of samples is defined as Equation ( 1), where 1(x) is the indicator function.
accuracyðy, b yÞ In the binary classification task, the terms "positive" and "negative" refer to the prediction of the classifier, and the terms "true" and "false" refer to whether that prediction corresponds to an external judgment.Given these definitions, we can formulate the following: t p (true positive), f p (false positive), f n (false negative), and t n (true negative).Intuitively, precision is the ability of a classifier not to label negative samples as positive, and recall is the ability of a classifier to find all positive samples.The F-measure (F β and F i measures) can be interpreted as a weighted harmonic average of precision and recall.When β = 1, F β and F i are equivalent.We have to define equations as follows: 3. Results

Analysis of the Relationship between Gait Characteristics and Personality
We performed dimensionality reduction on all gait features obtained from the algorithm to reduce their complexity, enhance interpretability, and facilitate visualization.Specifically, we initially computed the correlation coefficients between gait features and personality values using the correlation coefficient method, and selected features with correlation coefficients greater than 0.3.Subsequently, we applied the random forest feature importance ranking algorithm to rank the selected features from the previous step.We then compared and analyzed the prediction factors that contributed over 90% to the four dimensions, and presented the results in Figure 9.We combined the gait data and personality information to explore the correlation between walk and personality.We also tried to uncover the connection behind the correlation.By visualizing the quantitative data (Figure 9), we systematically analyzed and explained the differences between the four dimensions of E-I, S-N, T-F, and J-P between the groups.We believe that identifying the similarities and differences between the groups of personality preferences can help us understand the reasons behind the covariation between personality and behavior, specifically, which gait parameters can explain the similarities and differences in the groups and how such similarities and differences can be explained.We believe this is necessary to advance our understanding of personality to a considerable extent.In order to make the analysis of gait and personality more comprehensive, we have added an analysis of the relationship between gait and personality for different genders.This is because gender is an important factor that affects gait and personality analysis.

E-I Dimension
The E-I dimension describes the way an individual directs his/her energy.Extroverts feel energized when spending time with others or in busy, active environments.In addition, extroverts tend to be expressive and outspoken.By contrast, introverts feel energized by spending quiet time alone or with a small group.Introverts also tend to be conservative and considerate. [41]e found that the difference between the two types in the process of gait movement was obvious in the parameters, as shown in Figure 9a.The range distribution of angular velocity in the forward direction of the extroverts was wider and more diffused than that of the introverts, and most were higher than those of the introverts.The acceleration data distribution of the introverts in the forward direction was not as wide as that of the extroverts, and the data were relatively concentrated at the large value, thereby indicating that they were more focused on walking compared with the extroverts.The variance of the speed of the extroverts in the Y-axis direction was higher than that of the introverts on average and more dispersed, indicating that their walking fluctuated considerably.Moreover, the acceleration range in the vertical direction of the extroverts was wider and higher than that of the introverts.Judging from the angular velocity in the Y-axis direction, we determined that the introverts were mostly higher than the extroverts, and the data changed within a small range.The stride frequency of the introverts was higher than that of the extroverts, and the data were similar, and the distribution was more concentrated, thereby indicating that the introverts have faster and shorter strides.Our comprehensive findings determined the following: the individuals who walked briskly, with a large range, and wantonly were extroverts, whereas those who walked cautiously, with a tight rhythm, and relatively calmly and attentively were introverts.This finding also confirmed the idea that introverts prefer to enjoy their time.
Based on the results of gender analysis of gait and personality presented in Figure 10 and 11, it was observed that female extroverts generally exhibit better social interaction and socializing skills, which may be reflected in their gait.They are more likely to display a wider roll angle, which could be associated with increased social interaction, observation, and reaction during walking.Extroverts may be more responsive to external stimuli, which could result in a larger roll angle while walking.In contrast, introverts tend to focus more on their own feelings and thoughts, thereby concentrating on maintaining a steady pace while walking.This results in a lower roll angle but higher lateral acceleration during their gait.In males, the median vertical acceleration and the standard deviation of acceleration in the forward direction of extroverted (type E) individuals were significantly lower than those of introverted (type I) individuals.This finding may reflect that extroverted individuals tend to exhibit more stable and direct behavioral traits, while introverted individuals may focus more on internal thinking and reflection, leading to greater variation and fluctuation in their actions.This observation emphasizes that an individual's personality traits can impact their behavior, which can even be observed during everyday activities.

S-N Dimension
The S-N dimension describes how an individual obtains information.Sensibles focus on their senses and are interested in information that can be seen, heard, felt, and so on directly.Intuitives  S, c) T-F, d) J-P.Note: "a" represents acceleration, "v" represents velocity, and "g" represents angular acceleration; statistical features include "range"range, "med"median, "var"sample variance, "Hm"harmonic mean, "min"minimum, "max"maximum, "std"standard deviation, "mean"average , "cadence"steps per second, "area" sum of amplitudes, "rms"root-mean-square value, and "roll"angle of counterclockwise rotation around the positive direction of the X-axis.
focus on abstract levels of thinking and are interested in theories, patterns, and explanations. [41]e found that the difference between the two types in the process of gait movement was obvious in the parameters, as shown in Figure 9b.From the figure, we can see that in the Y-axis direction, the variance of the angular velocity of the sensibles was higher than that of the intuitives on average, and the fluctuation range was large.In the forward direction, the acceleration of the intuitives was mostly higher than that of the sensibles, and the variance was large, that is, the data fluctuated considerably.The average speed of the intuitives was slow and concentrated, and they walked smoothly.In the vertical direction, the speed of the intuitives hastened, and their walking was stable.The roll value of the sensibles was higher than that of the intuitives, and the data were widely distributed.Regarding the explanation for this outcome, in walking movement, sensibles obtain cues from their environment through perception, whereas intuitives, who are alert to their surrounding environment, obtain cues through intuition.Overall, those who walked briskly and actively were sensibles, whereas those who walked steadily and demonstrated effective thinking abilities were intuitives.
According to the results, there were significant differences in walking patterns between N-type and S-type females.Specifically, N-type individuals exhibited a higher range of lateral angular velocity and a higher average forward direction while walking.This can be explained by the tendency of N-type individuals to focus on abstract thinking, future trends, and meanings, which is reflected in their more flexible and open walking style, making them more adaptable to different situations.In contrast, S-type individuals pay more attention to details and specific experiences, resulting in a more stable walking pattern that emphasizes  maintaining a straightforward direction.In males, S-type individuals exhibited a significantly lower range of vertical acceleration and maximum acceleration while walking compared to N-type individuals.This reflects the sensory orientation of individuals, with S-type individuals focusing more on the specific details and sensory experiences of their surroundings, leading to a greater emphasis on maintaining a stable and regular pace while walking, rather than focusing on abstract thinking and future trends like N-type individuals, resulting in a smaller range of vertical acceleration.

T-F Dimension
The T-F dimension describes how an individual makes decisions.T-people tend to make decisions using logic and are interested in finding the most logical and reasonable options.F-people tend to make decisions using personal values and are interested in how their decisions will affect other people and whether they align with their values. [41]e found that the difference between the two types in the process of gait movement was obvious in the parameters, as shown in Figure 9c.The area value of the T-people in the forward direction was mostly higher than that of the F-people and very concentrated, thereby indicating that their acceleration range was wider.The acceleration rms of the T-people in the Y-axis direction was large, and the data were scattered.In addition, the area value of the speed of the T-people was large in the vertical direction, and the data range was large.The mean value of the cadence of the F-people was slightly high, and the distribution was concentrated, thereby indicating that their walking rhythm was fast and close.Overall, the T-people walked steadily and cautiously, whereas the F-people walked tightly and quickly.
In our observations of female walking patterns, we found that T-type individuals exhibited significantly lower values in terms of maximum lateral angular velocity, while displaying significantly higher average vertical acceleration compared to F-type individuals.This can be attributed to the tendency of T-type individuals to prioritize logic and facts, focusing more on details and planning, resulting in a preference for stable and organized movements during walking.In contrast, F-type individuals place more emphasis on emotions and interpersonal relationships, leading to fewer dynamic changes in the vertical direction.In males, T-type individuals showed a significantly higher range of forward acceleration and maximum lateral acceleration compared to F-type individuals.This reflects the personality trait of T-type individuals, who prioritize logic and rationality, leading them to adopt a more direct and goal-oriented walking style, resulting in higher forward acceleration and lateral acceleration.In contrast, F-type individuals place more emphasis on emotions and values, leading to a preference for smooth and cautious strides while walking, resulting in lower acceleration values.

J-P Dimension
The J-P dimension describes how a person processes the structure of the world around him/her.J's appreciate structure and order and enjoy following a plan, whereas P's appreciate flexibility and spontaneity, enjoy being open, and can change their minds at any time. [41]e found that the difference between the two types in the process of gait movement was obvious in the parameters, as shown in Figure 9d.From the figure, we can see that in the vertical direction, the average value of the area value of the speed of the J's was slightly higher than that of the P's, but the variation was small, thereby indicating that the velocity range in this direction was wider.The acceleration range of the J's in the forward direction was larger than that of the P's, and the distribution was more concentrated, thereby indicating that the acceleration fluctuation range of the J's when walking was wider.The harmonic mean and median of the angular velocity of the J's were small and concentrated in the Y-axis direction, which meant that their angular velocity was concentrated in a relatively small value.Overall, the J's walked relatively steadily, whereas the P's walked smoothly and freely, focusing on creativity and randomness.
When walking, females who are J-type individuals exhibit significantly lower values in terms of maximum vertical angular velocity and pitch angle integration area compared to P-type individuals.This can be attributed to the tendency of J-type individuals to prioritize organized and planned actions, as well as a stronger sense of responsibility.Therefore, J-type individuals tend to exhibit more cautious and controlled movements while walking, which is reflected in their lower vertical angular velocity and pitch angle integration area.In males, J-type individuals showed a significantly lower variance in forward acceleration and average vertical acceleration compared to P-type individuals.This reflects the characteristics of the two personality types.J-type individuals typically prioritize organized and planned actions, preferring stability and accuracy.Therefore, they exhibit a lower variance in forward acceleration while walking, helping to maintain a consistent pace.In contrast, P-type individuals are more open and adaptable, exhibiting a higher average vertical acceleration, reflecting a faster response to environmental changes.

Personality Measurement
The classification results of five classifiers are shown in Figure 12.Compared with other methods, random forest has the best performance in four classification dimensions: accuracy, precision, recall, and F1-score.So, we selected the random forest algorithm to predict the four MBTI preferences.
The results of the confusion matrix are presented in Figure 13a.Accuracy for the E-I dimension was 95%, accuracy for the S-N dimension was 96%, accuracy for the T-F dimension was 91%, and accuracy for the J-P dimension was 91%.All the average accuracy rates were higher than 90%.This accuracy exceeded current predictions based on features such as face, voice, or everyday behaviors to demonstrate the high interpretability of the gait parameters selected in this study, and their actual correlation with the 16 personality types.
Furthermore, we chose a cross-validation method to verify the reliability of the model and used the average value of the accuracy of the results as an estimate of the accuracy of the algorithm to describe the reliability of the experimental method.For the analysis of the experimental results, we selected metrics such as accuracy, precision, recall, and the F1-score (Table 4) to evaluate the performance of our trained models.
Based on the prediction results of the above four dimensions and our existing sample size, we predicted the 16 MBTI personality types.Owing to the uneven distribution of the proportion of each personality type and the intersection of the personality dimensions for each personality type, we used a boosting-based technique to create a robust learner that can make accurate predictions of the 16 personality types for the participants.The results from the questionnaire test were used as the true label, and the prediction accuracy of the 16 types achieved 88.16% (i.e., the percentage of all correctly predicted samples versus all samples).The prediction accuracy of each type can be found at the diagonal cell of Figure 13b, where the other numbers show how the sample was misclassified.

Discussion
Some previous studies confirmed the correlation between human walking patterns and psychological characteristics, including personality traits. [23,24,47]However, limited by   measurement methods for gait and personality or the experimental environment, the application of engineering technology methods in psychological research is uncommon, and reports on systematic and quantitative measurement and research on personality through gait behavior are few. [49]In this study, first, we determined the subjects' 16 personality types through standard MBTI assessment and obtained and measured their ankle movement data using a self-developed wearable sensor.Second, we solved the gait information parameters via coding to quantify the correlation between gait and personality with a machine learning algorithm and systematically described the specific relationship between the four dimensions of the MBTI personality types and gait behavior.In this study, we found that gait had a high degree of explanation for the preferences of the E-I and S-N dimensions, with an accuracy rate of over 95%.Specifically, the extroverts walked briskly, with a considerable range, and freely, giving the impression that they were very energetic.By contrast, the introverts tended to be slightly restrained and tight paced when walking as well as relatively calm and focused.The sensibles walked briskly and regularly and were accustomed to obtaining clues from the environment through perception during the walking process.Meanwhile, the intuitives had effective thinking skills and walked steadily.The accuracy rate of the gait parameters in describing the preferences of the T-F and J-P dimensions was above 90%, which was slightly lower than that of the gait parameters in describing the E-I and S-N dimensions.The T-people walked steadily and cautiously and tended to be effective thinkers, whereas the F-people walked tightly and hastily.Furthermore, the P's were creative and casual in their walking movements, walking relatively actively, loosely, and adventurously, whereas the J's walked relatively steadily.These findings can extend research on the correlation between gait and personality, and some are consistent with those of previous studies. [21]n addition, we verified the timeliness and stability of the gait-personality relationship model, and the results showed that gait was relatively stable for at least one year.The personality comparisons also met the standard test-retest level; thus, the results of both assessments were positive.However, we also noticed some limitations of our study.Although our proposed method is relatively objective, and the measurement accuracy obtained for the 16 personality types is as high as 88.16%.The results are higher than those obtained by other current methods of personality research, as shown in Table 5.The model may have errors due to the uncertainty or instability of human personality type variation over time (i.e., the reference data for personality types determination using MBTI may have errors over time).Among such errors, the low accuracy for the ISTJ, ESFJ, and ISFP types may be due to "sample misjudgment" owing to changes in certain personality dimensions during the retest.Some small changes in the model can make a big difference in the results, which is especially evident in personality.Therefore, future research can expand the sample size to examine more subjects (different ages, cultural differences, geographical differences, and so on) and use other advanced AI algorithms to evaluate the validity of gait movement for the personality measures.

Conclusion
The walking styles of people of different personality types show differences and details in gait features.Currently, vision-or voice-based personality prediction systems have an accuracy rate of around 80% or less.In this study, we introduce a system for MBTI personality measurement through gait using a single wireless IoT wearable motion sensor.The system collects human gait movement data and uses them for the accurate measurement of the four dimensions of personality (E-I, S-N, T-F, and J-P) and 16 corresponding personality types (ISTJ, ISFJ, INFJ, INTJ, ISTP, ISFP, INFP, INTP, ESTP, ESFP, ENFP, ENTP, ESTJ, ESFJ, ENFJ, and ENTJ).We extracted 170 gait parameter features based on an optimized algorithm from the ankle motion data and determined the most significant features to describe the difference in the gait of the four dimensions of personality.To perform the binary classification in each personality dimension, we tested a variety of machine learning algorithms (decision tree, logistic regression, support vector machines, random forest, and Naive Bayes) to find the model with the highest accuracy.Based on the experimental data, we observe that the random forest algorithm demonstrates the best performance and obtains results with a prediction accuracy of more than 90% for each of the four MBTI dimensions.Finally, we use a boosting-based learner to predict the 16 personality types and obtain a measurement accuracy of 88.16%.Beyan et al. [50] Big Five Video CNN þ LSTM 77% Marouf et al. [51] Big Five Text NB, RF, DT, SLR, SVM 61.89-72.13% Mawalim et al. [52] Big Five Multimodal RF 63%-70% Our work MBTI IMU gait data NB, RF, DT, LR, SVM 88.16%

Figure 1 .
Figure 1.MBTI 16 personality composition and structural chart.In the innermost circle are the four worldviews, each with two dimensions, the outer colorful circles show the 16 personality types.

Figure 2 .
Figure2.Radar chart comparing this study and other related studies in various dimensions.Note: "Analysis method" represents the depth of the method in the study, which is accumulated sequentially from the inside to the outside.

Figure 3 .
Figure 3. Overall system diagram.Part A shows the experiment and hardware diagram; self-developed MEMS IoT sensor is on the left, right side is the simulated experimental environment.Part B presents the data algorithm analysis, left side shows the main differences in gait characteristics of the four personality dimensions, and the right side shows the algorithm structure.Part C is the prediction results of the algorithm, the lower part is the prediction results of the four personality dimensions, and further corresponds to the above 16 personality measurements.

Figure 4 .
Figure 4. Experimental setup and data map.a) Experimental environment and the custom-built sensing device.b) The corresponding gait data graph while walking.

Figure 5 .
Figure 5. Preprocessed data corresponding to 16 personality types; the upper solid line is the acceleration data, and the lower dotted line is the angular velocity data.Note: The symbol "a" represents acceleration, and "σ" represents angular velocity.The unit of acceleration g is 9.8 m s À2 .

Figure 6 .
Figure 6.Distribution map of subjects' personality types.a) Distribution and proportion of personality types.b) Preference distribution of all subjects in four dimensions.

Figure 7 .
Figure 7. Gait stability analysis.Comparison of gait data after preprocessing, data of one gait cycle of subjects with 16 personality types before and after a year, the shaded area is the gait fluctuation range before and after one year.Note: The unit of acceleration g is 9.8 m s À2 , and Dg represents the angular velocity.

Figure 8 .
Figure 8. Personality stability analysis.Graph of the relative stability of personality types.The vertical axis in the graph represents the number of people.

Figure 10 .
Figure 10.Comparison of gait differences in four dimensions of female personality.

Figure 11 .
Figure 11.Comparison of gait differences in four dimensions of male personality.

Figure 12 .
Figure 12.Comparison of classification results of personality preferences of different dimensions based on five machine learning models.

Figure 13 .
Figure 13.Confusion matrix for predicting four dimensions and 16 personality types of MBTI by gait parameters.The color bar is shown on the right.Note: The darker the color, the higher the accuracy rate; part without number marks has an accuracy rate of 0.

Table 1 .
Recent research work on the classification/recognition of various "human information" based on inertial measurement unit (IMU) sensors.

Table 2 .
Summary information of the test subjects.

Table 3 .
Gait parameters used in this study.

Table 4 .
Performance index evaluation of optimal results.

Table 5 .
Comparison with results from other current personality research methods.