A Machine Learning Approach to Select the Type of Intermittent
Fasting in Order to Improve Health by Effects on Type 2 Diabetes
Shula Shazman
Department of Mathematics and Computer Science, The Open University, Derech Hauniversita 1, Raanana, Israel
Keywords: Machine Learning, Decision Tree, Type 2 Diabetes, Insulin Resistance, Precision Medicine, Intermittent
Fasting, Alternate Day Fasting.
Abstract: Intermittent fasting (IF) is the cycling between periods of eating and fasting. The main types of IF are:
complete alternate-day fasting; time-restricted feeding (eating within specific time frames such as the most
prevalent 16:8 fast, with 16 hours of fasting and 8 hours for eating); religious fasting such as the Ramadan
(occurs one month per year, with eating taking place only after nightfall). IF can be effective in reducing
metabolic disorders and age-related diseases by bringing about changes in metabolic parameters associated
with type 2 diabetes. Questions do remain, however, about the effects of the different types of IF as a function
of the age at which fasting begins, gender and severity of type 2 diabetes. In this paper we describe a machine
learning approach to selecting the best type of IF to improve health in type 2 diabetes. For the purposes of
this research, the health outcomes of interest are changes in fasting glucose and insulin. The different types
of intermittent fast offer promising non-pharmacological approaches to improving health at the population
level, with multiple public health benefits.
1 INTRODUCTION
Diabetes has become prevalent with changes in
lifestyle, threatening to reduce life expectancy for
humans around the globe. According to the
International Diabetes Federation (IDF), there were
425 million people in the world with diabetes in 2017
– close to 1 in 11 people (Diabetes Atlas 8
th
edition,
2017).
There are two main types of diabetes – type 1 and
type 2, both of which can lead to chronically high
blood sugar levels. People with type 1 diabetes barely
produce insulin at all, while those with type 2 diabetes
produce insulin but do not respond to it as they
should. Ninety to ninety-five percent of people living
with diabetes have type 2 diabetes.
Type 2 diabetes is generally characterized
by insulin resistance (IR), where the body does not
fully respond to insulin. IR is now used as a screening
index for primary prevention of type 2 diabetes.
Using the Homeostatic Model Assessment of Insulin
Resistance (HOMA-IR) equation, IR can be
estimated from fasting glucose and insulin levels. A
high score of HOMA-IR indicates significant insulin
resistance usually found in people with type 2
diabetes (Tang et al. 2015; Sharma and Fleming
2012).
Despite the awareness of the need for early
diagnosis, prevention and treatment of diabetes, the
IDF estimates that there will be 642 million people
living with the disease by 2040, and another half as
many who will be living with undiagnosed diabetes,
at unknowing risk of its disabling, life-threatening
complications (Diabetes Atlas 8
th
edition, 2017).
The cornerstone of type 2 diabetes management is
a healthy diet, increased physical activity and
maintaining healthy body weight. Oral medication
and insulin are also frequently prescribed to help
control blood glucose levels. A new precision
medicine approach is also necessary for treatment of
diabetes in addition to traditional approaches.
Daily calorie restriction regimens are still the
most common diet strategies implemented for
improving HOMA-IR (Wilding 2014). These are
effective for weight loss in some individuals, but
many people find this type of diet difficult, as it
requires vigilant calorie counting on a daily basis, and
the sense of never being able to eat freely throughout
the day results in dieter frustration.
These impediments to the calorie restriction
approach have brought about the introduction of
Shazman, S.
A Machine Learning Approach to Select the Type of Intermittent Fasting in Order to Improve Health by Effects on Type 2 Diabetes.
DOI: 10.5220/0008950201310137
In Proceedings of the 13th International Joint Conference on Biomedical Engineering Systems and Technologies (BIOSTEC 2020) - Volume 3: BIOINFORMATICS, pages 131-137
ISBN: 978-989-758-398-8; ISSN: 2184-4305
Copyright
c
2022 by SCITEPRESS Science and Technology Publications, Lda. All rights reserved
131
another approach termed intermittent fasting (IF),
which has proven promising in achieving reduction in
HOMA-IR, although not in all cases. IF is a form of
time-restricted eating; it differs from calorie
restriction in that the individual is only required to
restrict energy intake during a portion of the day
(typically 16 hours), and allows for free food
consumption in the non-restricted hours. Alternate
day fasting is a subclass of IF, consisting of a ‘fast
day’ alternating with a ‘feed day’ (ad libitum, which
is eating food as much as desired).
Previous studies and reviews provide an overview
of IF regimens (Patterson and Sears, 2017;
Malinowski et al., 2019; Ganesan et al., 2018;
Barnosky et al., 2014), and summarize the evidence
for their health benefits. Furthermore, they discuss
physiological mechanisms by which IF might lead to
improved health outcomes. They have not provided a
clear answer, however, to the question of whether IF
is always able to reduce HOMA-IR; that is, the
conditions (age, gender, basal fasting glucose level,
etc.) needed to make the IF effective for reducing
HOMA-IR have not yet been deciphered. Moreover,
no previous IF study has reported results per
individual; results were reported on a group level
only.
In today's era of precision medicine, the current
study has been motivated to answer the question of
whether a patient with prediabetes or diabetes could
benefit from an intervention, reducing HOMA-IR or
even eliminating the disease altogether. This study
suggests a recommendation system based on
individual data from human fasting intervention
studies, where the health outcomes of interest are
changes in metabolic parameters associated with type
2 diabetes. The system presented, based on a
machine-learning approach, predicts which type of IF
treatment can improve an individual's health by
reducing insulin resistance and preventing or curing
type 2 diabetes.
The results of this study provide a set of rules
which can assist individual patients and their
physicians in selecting the best IF intervention for
their personal case.
2 METHODS
This study aims to predict whether a specific IF
intervention would reduce the insulin resistance of an
individual with prediabetes. The approach contains
four basic steps: identifying required data, preparing
and pre-processing, modeling the data and finally,
training and testing.
2.1 Identifying Required Data
In order to answer the question of this study I asked
for the individual data from authors of 25 published
papers that performed randomized clinical trials
investigating the IF effects on type 2 diabetes
parameters. I received the individual data from 6 out
25 papers (Halberg et al., 2005; Harvie et al., 2011;
Harvie et al., 2013; Clifton et al., 2004; Chowdhury
et al., 2016a; Chowdhury et al., 2016b). The rest of
authors responded that they could not send the data
due to participant confidentiality.
2.2 Preparing and Pre-processing the
Data
2.2.1 Selecting Individuals
From all the data received, 254 individuals with basal
fasting glucose above 5 mmol/L (90 mg/dL) or BMI
(Body Mass Index) above or equal to 25 were
selected. The selection criteria were established since
they indicate possible prediabetes (IDF Diabetes
Care. Volume 42, Supplement 1, January 2019). The
IDF's 2019 cutoff for fasting glucose indicating
prediabetes is 100 mg/dL; we set the cutoff at 90
mg/dL. (The table containing the data may be found
in Supplement 1 at the following link: https://
github.com/shulash3/intermmitentFasting/blob/
master/Supplementary1.xlsx).
2.2.2 Calculating HOMA-IR
The Homeostatic Model Assessment of Insulin
Resistance (HOMA-IR) has been proven to be a very
sensitive test for indicating prediabetes (Sharma and
Fleming 2012). Using the HOMA-IR equation,
insulin resistance can be estimated from fasting
glucose and insulin levels.
HOMA-IR = Fasting Glucose * Fasting Insulin
(1)
A high score of HOMA-IR indicates significant
insulin resistance, usually found in people with type
2 diabetes.
For each of the 254 individuals, we calculated the
HOMA-IR twice using Equation 1 – once for the
basal values of fasting glucose and insulin and once
for the values after the intervention. The difference
between them represents the insulin resistance
reduction.
2.2.3 Intermittent Fasting Interventions
The dataset included 9 different types of
interventions, e.g. continuous energy restriction – a
BIOINFORMATICS 2020 - 11th International Conference on Bioinformatics Models, Methods and Algorithms
132
seven day-a-week trial; intermittent energy restriction
– a two day-a-week trial allowing eating freely in the
remaining 5 days; daily morning fasting; or fasting
every second day. Part of the interventions contained
specific diets such as carbohydrate restriction; high
carbohydrate; or high monounsaturated.
Table 1: IF regimens.
Intervention
name
Details Reference
CER
Continuous Energy Restriction
– 7-day-a-week trial; eating
restricted calories every day.
Harvie et al.
2011
IER
Intermittent Energy Restriction
2-day-a-week trial; eating
restricted calories only two
days a week.
Harvie et al.
2011
DMF
Daily Morning Fasting; start
eating at noon and finish at
20:00.
Chowdhury et
al. 2016a and
Chowdhury et
al. 2016b
FESD
Fasting Every Second Day;
eating only four days a week.
Halberg et al.
2005
IECR
Intermittent Energy and
Carbohydrate Restriction;
eating restricted calories only
two days a week.
Harvie et al.
2013
IECR+PF
Intermittent Energy and
Carbohydrate Restriction +
free Protein and Fat; eating
restricted calories only two
days a week.
Harvie et al.
2013
DER
Daily Energy Restriction;
eating restricted calories every
day.
Harvie et al.
2013
High Carb
High Carbohydrate weight
loss diet; eating restricted
calories every day.
Clifton et al.
2004
High Mono
High Monounsaturated weight
loss diet; eating restricted
calories every day.
Clifton et al.
2004
Table 1 summarizes the different IF regimens
included in this study. The reference to each regimen
is shown in the table for further details.
2.2.4 Selecting the Features
The initial vector of features for every individual is
shown in Figure 1A. The vector is composed of
details regarding the individual (age, gender, weight,
ethnicity, basal BMI, basal fasting glucose, fasting
glucose after intervention, basal fasting insulin and
fasting insulin after intervention) and details
regarding the intervention (intervention name and
duration).
Figure 1B describes the training vector, which is
the vector after removing the features 'fasting glucose
after intervention' and 'fasting insulin after
intervention'. The calculation of the HOMA-IR
difference is added to the training vector as follows:
if the intervention is successful we expect a reduction
in HOMA-IR; thus, if the HOMA-IR difference is
greater than zero the assignment in the 'HOMA-IR
difference' column is set to TRUE otherwise it is
FALSE.
Figure 1: Vectors of initial and training features.
2.3 Modeling the Data
Data mining tools such as classification, clustering,
association and neural networks solve problems by
analyzing large volumes of data. Classification is
possibly the most frequently used data mining
technique. In this study we address a classification
problem. Classification is the process of finding a set
of models that describes and differentiates data
classes and concepts, for the purpose of being able to
use the model to predict the class whose label is
unknown. There are many algorithms that can be used
for classification, e.g. decision trees, neural networks,
logistic regression and others. However, the decision
tree classification with the Waikato Environment for
Knowledge Analysis (Weka) is the simplest way to
mine information from a database. Furthermore,
decision trees are a way of representing a sequence of
rules that lead to a class or value. A decision tree is a
flowchart-like tree structure.
The decision tree algorithms J48, LMT (Logistic
Model Tree), Random Forest and Random Tree as
well as the Logistic Regression and Naïve Bayes
classifiers were tested on the data in this study.
2.4 Training and Testing
The 254 samples in the training data were trained by
the J48 decision tree (Weka 3.8.3). The
implementation of the J48 decision tree in Weka 3.8.3
can handle categorical and numerical attributes like
those found in our mixed dataset (Sewaiwar and
A Machine Learning Approach to Select the Type of Intermittent Fasting in Order to Improve Health by Effects on Type 2 Diabetes
133
Verma 2015). The optimal number of features as a
function of sample size is proportional to
for
highly correlated features (Hua et al. 2004). The
features in the study shown here are highly correlated;
254
15.9 while the number of features is 9 (i.e. 9
attributes for 254 individuals is reliable).
Two test approaches were selected to validate the
model – the Leave-One-Out and the 10-Fold cross-
validations.
3 RESULTS
3.1 HOMA-IR Reduction
All results of the six different classifiers – J48, LMT,
Random Forest, Random Tree, Logistic Regression
and Naïve Bayes – are shown in Table 2. The Area
Under the Curve (AUC) of the10-Fold test is shown
in the first row of the table while the second row
contains the data of the Leave-One-Out test. The
AUC values of the 10-Fold test range between 0.67
and 0.75 while those of the Leave-One-Out range
between 0.65-0.80. For both tests the AUC ranges are
very small; we therefore conclude that for this case all
six classifiers perform similarly. Finally, the J48
(C4.5) decision tree (Weka 3.8.3) is selected to model
the data of this study. Although the advantage of
Random Forest is to prevent overfitting by creating
random subsets of the features and building smaller
trees and then combining the subtrees, J48 is found to
produce the most accurate prediction among the
decision tree algorithms (Sewaiwar and Verma 2015).
Furthermore, J48 is self-explanatory and easy to
follow. The J48 decision tree is a predictive machine-
learning model which selects a target value (HOMA-
IR difference TRUE or FALSE) of an individual and
an intervention based on the training vectors
available. In the J48 decision tree, the different
features (age, gender, weight, etc.) are denoted by the
internal nodes of a decision tree, the branches
between the nodes tell us the possible values that
these features may have in the experimental samples
(gender: male/female, etc.), while the terminal nodes
tell us the final value of the dependent variable
(TRUE or FALSE assigned for HOMA-IR
difference).
The result of testing the J48 decision tree's model
using the 10-Fold cross validation test show that the
model predicts correctly in 72% of the cases, and the
AUC is 0.7. Furthermore, the Leave-One-Out test
achieves 78% accuracy and an AUC of 0.8. The
results suggest that the model can predict correctly in
78% of the cases whether an intervention would help
an individual improve their type 2 diabetes risk
parameters by reducing HOMA-IR.
Table 2: AUC for different classifiers.
J48 LMT
Random
Forest
Random
Tree
Logistic
Naive
Bayes
10-Fold 0.7 0.75 0.75 0.67 0.79 0.73
Leave-One-
Out
0.8 0.74 0.74 0.66 0.79 0.72
The visualization of the complete J48 decision
tree is found in Figure 2A, with detailed views shown
in Figures 2B, 2C, 2D and 2E.
A larger figure of 2A can be found in Supplement
2, at the following link: https://github.com/shulash3/
intermmitentFasting/blob/master/Supplementary2.pn
g. In Figure 2A the relative positions of Figures 2B,
2C, 2D, 2E are visible.
Figure 2A: Visualization of complete J48 decision tree.
The first node in the tree, as shown in Figure 2A,
is the gender feature, indicating that this attribute is
the most informative one for the decision.
Interestingly we also notice in Figure 2A that for
males the most important feature in determining
whether an intervention would be effective is the fast
duration while for females the basal fasting insulin
level is reported as the most important feature. In
figures 2B-2E TRUE (colored green) indicates
success in reducing HOMA-IR while FALSE
(colored red) indicates no reduction.
BIOINFORMATICS 2020 - 11th International Conference on Bioinformatics Models, Methods and Algorithms
134
Figure 2B: Male sub
-
decision tree.
It can be observed from Figure 2B that men are
indifferent to any of the intervention types, but the
duration of the intervention plays an important role.
Short duration of fasting and lower BMI or long
duration of intervention and younger age lead to the
success of the intervention (reducing HOMA-IR).
Reasonably, attributes like lower BMI and younger
age make it easier to reduce HOMA-IR.
Figure 2C: Female sub
-
decision tree.
In Figure 2C the different interventions appear to
be part of the decision nodes. The interventions are
colored yellow while the features are light brown.
The tree view on the female side is more complex.
This may be because there are more women in the
dataset than men. In Figure 2C the different
interventions appear to be part of the decision nodes.
These are organized hierarchically beginning with
DMF followed by IECR or beginning with IECR
followed by the Hi Mono diet.
In Figure 2D we see a hierarchical
structure of the
interventions ordered by their success in improving
HOMA-IR, beginning with DMF, IECR and then
IECR+PF.
Interestingly in Figure 2E there is a node where
lower BMI leads to an unsuccessful intervention. This
evidence should be further investigated.
Figure 2D: Female left sub
-
decision tree.
Figure 2E: Female right sub
-
decision tree.
3.2 Fasting Glucose or Fasting Insulin
Reduction
Prediction results of fasting glucose reduction and
fasting insulin reduction taken separately instead of
HOMA-IR reduction are shown in Table 3.
Table 3: Summary of AUC results for improving type 2
diabetes risk parameters.
HOMA-IR
reduction
FASTING
Glucose
reduction
FASTING
Insulin
reduction
10-Fold Cross
Validation
test
0.7 0.6 0.55
Leave- One-
Out test
0.8 0.6 0.6
A Machine Learning Approach to Select the Type of Intermittent Fasting in Order to Improve Health by Effects on Type 2 Diabetes
135
The results in Table 3 show that the HOMA-IR
improvement prediction is more effective than the
prediction of the fasting glucose or the fasting insulin
reduction taken separately. As shown in Equation 1,
the HOMA-IR calculation is based on both fasting
glucose and fasting insulin.
3.3 Random Classification
In order to validate that these results for HOMA-IR
cannot be achieved randomly, I reordered the values
in the HOMA-IR column in an arbitrary way. The
proportion between the TRUE values and the FALSE
values remained the same as in the original column.
Then I trained and tested the data once more. The
results of the random tests were much lower in AUC
compared with the original data. The results for the
10-Fold cross validation test were 0.56 AUC
compared with 0.7 in the original data. The results of
the Leave-One-Out test were even more significant –
0.61 AUC compared with 0.8 in the original data.
Those results suggest that the model predictions
cannot be achieved randomly.
3.4 Feature Selection
An interesting question is whether all the features
shown in Figure 1B are needed for the prediction. To
test this a feature selection test was performed on the
data. In each test a different feature was excluded.
The AUC results are shown in Table 4.
Table 4: Features selection – AUC results of J48 Decision
tree.
Excluded Feature
10-Fold Cross
Validation test
Leave-One-Out
test
None 0.7 0.8
Age 0.68 0.7
Gender 0.68 0.62
Weight 0.64 0.73
Ethnic 0.68 0.74
Basal BMI 0.69 0.77
Basal Fasting
Glucose
0.65 0.73
Basal Fasting
Insulin
0.62 0.6
The feature in every row of Table 4 other than the
first, is excluded and AUC is calculated without this
feature. None of the features is redundant, since the
higher AUC is shown when all features are trained.
Furthermore, J48 training and testing with data that
does not have more than one feature (from the list of
all nine features) resulted in even lower AUC values
than the values shown in Table 4.
4 CONCLUSIONS
Even a single fasting interval in humans (e.g.,
overnight) can reduce basal concentrations of
metabolic biomarkers such as insulin and glucose,
associated with chronic disease. For example,
patients are required to fast for 8–12 hours before
blood draws to achieve steady-state fasting levels for
many metabolic substrates. Recent studies suggest
that intermittent fasting regimens may be a promising
approach to losing weight and improving metabolic
health for people who can safely tolerate intervals of
non-eating, or eating very little, for certain hours of
the day, night, or days of the week. Furthermore,
these eating regimens may offer promising non-
pharmacological approaches to improving health at
the population level with multiple public health
benefits.
This study does not investigate weight loss;
however, it offers a recommendation system based
on data from several clinical trials for selecting the
optimal intervention to improve the health of
prediabetes individuals by reducing their type 2
diabetes risk parameters. The procedure in this study
is built using a machine learning approach and is
represented by a decision tree. The decision rules
derived from the tree are shown in Figures 2B-2E and
in the figure in Supplement 2 (which contains the
entire picture of decision rules). First, we observe that
males and females have a different set of rules, since
the node gender comes first in the tree. Males are
indifferent to the type of intervention; the success of
the intervention in males, however, is dependent on
IF duration. For example, if the duration of the
intervention is less than or equal to 2.5 weeks than the
success of the intervention depends on BMI. Males
with a smaller BMI will be more likely to have a
successful intervention. On the other hand, if the
duration of the intervention is more than 2.5 weeks
for males than age will be important to its success.
Reasonably, younger age will serve as a benefit. As
for females, most important for a successful
intervention is the level of basal fasting insulin. In the
case of a female with a basal fasting of less than or
equal to 37.1 pmol/L (for moderate insulin resistance
the fasting insulin should be in the range of 18–48
pmol/L) and age exceeding 52, there is no
intervention in the dataset that can assist in improving
HOMA-IR. Additional data from clinical trials can be
useful for expanding the recommendation system and
BIOINFORMATICS 2020 - 11th International Conference on Bioinformatics Models, Methods and Algorithms
136
applying it to a wider population. Furthermore, a
wider dataset will make if possible, to answer a more
interesting question, which is to predict what the best
fasting approach would be considering one's age,
gender, etc.
An algorithm which would answer the
above question would certainly assist physicians in
providing personalized medical advice to their
patients.
ACKNOWLEDGEMENTS
I wish to thank Michelle Harvie, Nils Halberg,
Flemming Dela, Peter M. Clifton, Eric Ravussin,
Leonie Kaye Heilbronn and Enhad Chowdhury for
their assistance in this study by sending the individual
data from their published clinical trial papers.
REFERENCES
Barnosky, A. R., Hoddy, K. K., Unterman, T. G. and
Varady, K. A., 2014. Intermittent fasting vs daily
calorie restriction for type 2 diabetes prevention: a
review of human findings. Translational Research
Volume 164, Number 4.
Chowdhury, E. A., Richardson, J. D., Holman, G. D.,
Tsintzas, K., Thompson, D. and Betts, J. A. 2016. The
causal role of breakfast in energy balance and health: a
randomized controlled trial in obese adults.Am J Clin
Nutr 2016;103:747–56. Printed in USA.
Chowdhury, E. A., Richardson, J. D., Tsintzas, K.,
Thompson, D. and Betts, J.A. 2016. Effect of extended
morning fasting upon ad libitum lunch intake and
associated metabolic and hormonal responses in obese
adults. International Journal of Obesity (2016) 40,
305–311Macmillan Publishers Limited.
Clifton, P. M, Noakes, M. and Keogh, J. B. 2004. Very
Low-Fat (12%) and High Monounsaturated Fat
(35%) Diets Do Not Differentially Affect Abdominal
Fat Loss in Overweight, Nondiabetic Women.
Human Nutrition and Metabolism.
Ganesan, K., Habboush, Y. and Sultan, S. 2018.
Intermittent Fasting: The Choice for a Healthier
Lifestyle. Cureus 10(7): e2947. DOI
10.7759/cureus.2947.
Halberg, N., Henriksen, M., So¨derhamn, N., Stallknecht,
B., Ploug, T., Schjerling, P. and Dela, F. 2005. Effect
of intermittent fasting and refeeding on insulin action in
healthy men. J Appl Physiol • VOL 99 • DECEMBER
2005.
Harvie, M. N., Pegington, M., Mattson, M. P., Frystyk, J.,
Dillon, B., Evans,G., Cuzick,J., Jebb, S. A., Martin, B.,
Cutler, R. G., Son, T. G., Maudsley, S., Carlson, O. D.,
Egan, J. M., Flyvbjerg, A. and Howell, A. 2011. The
effects of intermittent or continuous energy restriction
on weight loss and metabolic disease risk markers: a
randomized trial in young overweight women.
International Journal of Obesity (2011) 35, 714–727
Macmillan Publishers Limited.
Harvie, M. N., Wright, C., Pegington, M., McMullan, D.,
Mitchell, E., Martin, B., Cutler, R. G., Evans, G.,
Whiteside, S., Maudsley, S., Camandola, S., Wang, R.,
Carlson, O. D., Egan, J. M., Mattson, M. P. and Howell,
A. 2013. The effect of intermittent energy and
carbohydrate restriction v. daily energy restriction on
weight loss and metabolic disease risk markers in
overweight women. British Journal of Nutrition (2013),
110, 1534–1547.
Hua, J., Xiong, Z., Lowey, J., Suh, E. and Dougherty, E. R.
2004. Optimal number of features as a function of
sample size for various classification rules.
Bioinformatics.
International Diabetes Federation 2017. Diabetes Atlas 8
th
edition.
International Diabetes Federation. Diabetes Care. Volume
42, Supplement 1, January 2019
Malinowski, B., Zalewska, K., W˛esierska, A.,
Sokołowska, M. M., Socha, M., Liczner, G., Pawlak-
Osi´nska, K. and Wici´nski, M. 2019. Intermittent
Fasting in Cardiovascular Disorders-An Overview.
Nutrients.
Patterson, R. E. and Sears, D. D. 2017. Metabolic Effects
of Intermittent Fasting. Annual Review of Nutrition.
Sewaiwar, P. and Verma, K.K. 2015. Comparative Study of
Various Decision Tree Classification Algorithm Using
WEKA. International Journal of Emerging Research in
Management &Technology ISSN: 2278-9359 (Volume-
4, Issue-10).
Sharma, S. and Fleming, E. 2012. Use of HbA1C testing to
diagnose pre-diabetes in high risk African American
children: A comparison with fasting glucose and
HOMA-IR. Diabetes & Metabolic Syndrome: Clinical
Research & Reviews 6 (2012) 157–162.
Tang, Q., Li, X., Song, P. and Xu. L. 2015. Optimal cut-off
values for the homeostasis model assessment of insulin
resistance (HOMA-IR) and pre-diabetes screening:
Developments in research and prospects for the future.
Drug Discoveries & Therapeutics. 2015; 9(6):380-385.
Wilding, J. P. 2014. The importance of weight management
in type 2 diabetes mellitus. The Authors International
Journal of Clinical Practice Published by John Wiley
& Sons Ltd.
A Machine Learning Approach to Select the Type of Intermittent Fasting in Order to Improve Health by Effects on Type 2 Diabetes
137