# RESULTS AND DISCUSSION

6.6.1 FOR ASIAN COUNTRIES

Three combinations of parameters are used for prediction using different algorithms for different Asian countries such that accuracy, precision, sensitivity, and specificity are calculated. Sensitivity and specificity rates

FIGURE 6.3 Workflow of the consensus-based final prediction.

are needed to draw the ROC. Three combinations of parameters were used for prediction: Combination 1: age, gender, physical activity, family history, and WC; Combination 2: age, gender, physical activity, family history, WC, and BMI; and Combination 3: age, gender, physical activity, family history, and BMI. While developing the model with logistic regression, initially, Xand Y values are defined. X is the matrix that contains the attributes from the dataset and Y is the vector, based on which prediction can be done. X is defined as the corresponding p values of age, waist, physical activity, family history, and BMI. Y is the vector based on the outcome. Once the X and Y values are defined, split the X and Y values into corresponding training set and testing set. Here, sklearn splitting is done such that the random state value is set as zero. Once the classification of training set and testing set are over in the next stage, on the training set, we have to train a logistic regression model and fit the model on the X_train and T train. Once the model is fit, prediction based on the testing set, that is,X_test, should be carried out and calculate the accuracy such that in this dataset, the accuracy for the combinations 1-3 for the Asian countries is up to 0.8505, 0.9779, and 0.960, respectively. Once the model is built, now, the confusion matrix is created such that it will give the number of real and false prediction in the form of array. The confusion matrix for our dataset for combination 2 of Asian countries is shown in Figure 6.4, which indicates that the dimension is 2x2. For example, in the Indian system, real prediction values are 3153 and 122,587 (diagonal values) and inaccurate prediction values are 1846 and 1010. Similarly, the confusion matrix for different countries is calculated, which is fully used for the prediction.

Therefore, from the logistic regression model, the classification rate, precision, recall, sensitivity, and specificity are shown in Table 6.4 for combination 2. Similarly, values are identified for remaining combination. True negative rate is determined by the specificity, which defines the percentage of patients who are correctly identified as being healthy, so using a logistic regression model for combination 2, it is almost up to 63.07%, 70.94%, 69.38%, and 80.18%, respectively. True positive rate is determined by the sensitivity, which defines the percentage of patients who are correctly identified as being disease so using the logistic regression model for combination 2, it is almost up to 99.18%, 98.64%, 98.74%, and 98.54%, respectively, for the Asian countries. Therefore, similarly, these values are identified using four different algorithms for Asian countries, as shown in Tables 6.5-6.7.

FIGURE 6.4 Visualizing the confusion matrix.

TABLE 6.4 Accuracy, Sensitivity, and Specificity Percentage Using Machine Learning Algorithms

 Logistic Regression Accuracy (O/O) Precision (0/0) Recall (o/o) Sensitivity (0/0) Specificity (0/0) India 97.78 98.51 99.1 99.18 63.07 China 96.98 98.14 98.6 98.64 70.94 Sri Lanka 97.08 98.18 98.7 98.74 69.38 Oman 97.175 98.41 98.5 98.54 80.18

The optimum value (>95%) is considered as the high risk score for the diabetes that is detected based on the ROCs for India, China, Sri Lanka, and Oman, as shown in Figure 6.5.

TABLE 6.5 Combination 1 (Without Considering BMI)

 Accuracy Precision Sensitivity Specificity India Logistic Regression 85.05 87.40 91.39 71.17 Gaussian Bayes Model 78.4 78 79.0 77.0 Random Forest 71.1 69.5 71.9 69.3 Decision Tree 72.7 67 67.9 67.8 China Logistic Regression 93.98 92.97 72.17 70.81 Gaussian Bayes Model 95.30 96 96.0 95.0 Random Forest 96 95 95.0 92.0 Decision Tree 89.2 85.5 70.2 65.8 Sri Lanka Logistic Regression 94.37 94.28 74.19 61.53 Gaussian Bayes Model 94.50 89 94 92 Random Forest 94.0 88.2 94.8 91.7 Decision Tree 92.0 77.7 70.3 65.6 Oman Logistic Regression 92.55 90.19 73.18 69.72 Gaussian Bayes Model 92.62 86.7 93.0 89.9 Random Forest 93.0 86 93.0 89.0 Decision Tree 92.0 79.0 79.0 66.1

The Indian system is cross-validated using AUC and its value is 0.9873. The China system is cross-validated using AUC and its value is 0.9870. The Sri Lanka system is cross-validated using AUC and its value is 0.9879. The Oman system is cross-validated using AUC and its value is 0.9883.

There might be unpredictable and unknown connections between the factors in the dataset. It is critical to find and evaluate how many factors in the dataset are dependent on one another. This information can enable to more readily set up the information to meet the desires for machine learning calculations. Factors inside a dataset can be connected for a number of reasons. Relation could be true, neutral, or zero depends on the movement of the two variables. Association can similarly be neural or zero, inferring that the variables are insignificant. Relation between different features for the Indian system is shown in Figure 6.6. The graph is plotted according to the pair using the correlation feature, as shown in Figure 6.7. A similar approach is applied to all the remaining T2D systems of Asian countries and then analyzed.

 Accuracy Precision Sensitivity Specificity India Logistic Regression 97.78 98.51 99.18 63.07 Gaussian Bayes Model 96.06 92.00 96.12 94.3 Random Forest 96.0 92.0 96.0 94.0 Decision Tree 93.7 98.1 74.8 72.9 China Logistic Regression 96.98 98.14 98.64 70.94 Gaussian Bayes Model 94.7 89.23 94.80 91.0 Random Forest 94.0 89.0 94.0 91.0 Decision Tree 94.0 90.01 75.7 56.0 Sri Lanka Logistic Regression 97.08 98.18 98.74 69.38 Gaussian Bayes Model 97.01 97.0 97.0 97.0 Random Forest 99.9 99.8 98.7 98 Decision Tree 96.7 97.7 66.2 64.0 Oman Logistic Regression 97.175 98.141 98.54 80.18 Gaussian Bayes Model 95.90 96.28 96.0 95.0 Random Forest 96.0 96.0 96.0 95.0 Decision Tree 92.14 72.47 76.05 78.20

TABLE 6.7 Combination 3 (Without Considering WC)

 Accuracy Precision Sensitivity Specificity India Logistic Regression 96.0 95.8 96.2 94.5 Gaussian Bayes Model 96.24 96.0 96.0 94.0 Random Forest 99.0 99.0 99.2 98.0 Decision Tree 92.5 95.0 84.0 78.0 China Logistic Regression 96.24 97.51 98.52 60.74 Gaussian Bayes Model 95.50 95.25 96.0 94.0 Random Forest 98.0 98.0 98.0 98.0 Decision Tree 96.0 96.0 96.0 94.0 Sri Lanka Logistic Regression 96.52 97.78 98.46 64.07 Gaussian Bayes Model 95.47 92.5 91.0 89.5 Random Forest 98.0 97.8 96.4 92.2 Decision Tree 80.0 93.0 84.0 79.0 Oman Logistic Regression 85.8 91.5 85.8 81.0 Gaussian Bayes Model 94.48 94.0 94.0 93.0 Random Forest 98.0 98.0 98.0 98.0 Decision Tree 95.0 92.0 90.15 89.7

FIGURE 6.5 Optimum value using ROC.

FIGURE 6.6 Correlation between the features for the Indian system.

FIGURE 6.7 Pair plot according to correlation feature values.

Using the ROC curve, the cutoff value for the risk score can be identified. The ROC curves were plotted for the diabetes risk score, the sensitivity was plotted on the v-axis, and the false positive rate (1 - specificity) was plotted on the .v-axis. The more precise segregating the test, the more extreme the upward part of the ROC bend and the higher the zone under the bend (AUC). The optimum value is considered as the high risk score for the diabetes that is detected based on ROCs. ROC curves using different combinations for the Indian system are demonstrated in Figure 6.8. Therefore, it is observed that logistic regression performs better compared to all other algorithms. Since the AUC is constantly used to identify how well the test is performed between the two gatherings like if the value of AUC increases, which indicates, the better is the test. Therefore, the Indian system is validated using the AUC, and its value is 0.98. Similarly, China, Sri Lanka, and Oman systems are also validated, and the obtained results are 0.98, 0.97, and 0.94, respectively. A similar approach is applied for remaining countries.

FIGURE 6.8 ROC curves showing the performance of the diabetes risk score in predicting diabetes.

Similarly, about 514,384 samples are used for analysis. Optimization of diabetes data using /г-nearest neighbors (KNN) classifier, DT, RF, SVM, and GB model is compared as shown in Table 6.8.

TABLE 6.8 Analysis of Optimization of Different Machine Learning Algorithms

 Machine Learning Algorithm Accuracy— India Accuracy— China Accuracy— Sri Lanka Accuracy— Oman KNN Classifier Training set: 1.00 Test set: 1.00 Training set: 1.00 Test set: 1.00 Training set: 1.00 Test set: 1.00 Training set: 1.00 Test set: 1.00 Decision Tree Training set: 1.00 Test set: 1.00 Training set: 1.00 Test set: 1.00 Training set: 1.00 Test set: 1.00 Training set: 1.00 Test set: 1.00 Random Forest Training set: 1.00 Test set: 1.00 Training set: 1.00 Test set: 1.00 Training set: 0.94 Test set: 0.94 Training set: 0.93 Test set: 0.93 Support Vector Machine Training set: 0.99 Test set: 0.99 Training set: 0.95 Test set: 0.95 Training set: 0.965 Test set: 0.965 Training set: 0.95 Test set: 0.95 Gaussian Bayes Model Training set: 0.9611 Test set: 0.9608 Training set: 0.9404 Test set: 0.9414 Training set: 0.9443 Test set: 0.9450 Training set: 0.9261 Test set: 0.9262

By considering the simple five parameters, namely, age, family history of diabetes, WC, physical activity, and BMI, the system was developed. Quickly, the data for these factors were obtained by five inquiries and scores acquired for these elements, as shown in Table 6.9.

6.6.2 FOR EUROPEAN COUNTRIES

Similarly to explained in Section 6.4.1, here, three combinations of parameters are used for prediction using different algorithms for different European countries such that accuracy, precision, sensitivity, and specificity are calculated. Sensitivity and specificity rates are needed to draw the ROC. The confusion matrix for the European dataset for combination 2 is shown in Figure 6.9.

Therefore, from the logistic regression model, the classification rate, precision, recall, sensitivity, and specificity are shown in Table 6.10 for combination 2. Similarly, the values are identified for remaining combinations. True negative rate is determined by the specificity; therefore, using the logistic regression model for combination 2, it is almost up to 55.70%, 82.55%, 56.62%, and 55.83%, respectively. True positive rate is determined by the sensitivity, which defines the percentage of patients who are correctly identified as being disease; therefore, using the logistic regression model for combination 2, it is almost up to 97.70%, 90.35%, 93.07%, and 91.98%, respectively, for European countries. Therefore, similarly, these values are identified using four different algorithms for European countries, as shown in Tables 6.11-6.13.

TABLE 6.9 Diabetes Risk Score for Asian Countries

 Parameters Risk Score Age <35 0 35-49 22 >50 34 Obesity Waist Circumference Female <80 cm, Male <90 cm 0 Female 80-89 cm. Male 90-99 cm 11 Female >90 cm. Male >100 cm 20 Physical Activity Vigorous Exercise 0 Mild Exercise 13 No Exercise 18 Family History Two Nondiabetic Parents 0 Either Parent Having Diabetes 18 Both Parent Having Diabetes 29 BMI <25 0 25-29 11 30-34 16 >35 29 Maximum Score 130

Score >95: Very High Risk, 70-95:High Risk, 35-69: Medium Risk, <35: Low Risk

FIGURE 6.9 Visualizing the confusion matrix.

TABLE 6.10 Accuracy, Sensitivity, and Specificity Percentage Using Machine Learning Algorithms

 Logistic regression Accuracy (O/o) ' Precision (O/O) Recall («/о) Sensitivity (O/o) Specificity (O/O) Cambridge 84.42 87.36 93.13 97.70 55.70 Fiance 87.58 90.36 90.35 90.35 82.55 UK 84.52 87.49 93.07 93.07 56.62 Danish 82.69 85.76 91.98 91.98 55.83

The optimum value (80%) is considered as the high risk score for the diabetes that is detected based on the ROCs for Cambridge, France, UK, and Danish, as shown in Figure 6.10.

FIGURE 6.10 Optimum value using ROC.

TABLE 6.11 Combination 1 (Without Considering BMI)

 Accuracy (%) Precision (%) Sensitivity (%) Specificity (%) Cambridge Logistic Regression 78.87 82.06 97.89 92.66 Gaussian Bayes Model 83.54 83 84 81 Random Forest 78 75 78 75 Decision Tree 72 80 72 74 France Logistic Regression 73.94 78.76 81.55 60.17 Gaussian Bayes Model 71.3 73 71 72 Random Forest 77 74 76 73 Decision Tree 70 75 70 71 UK Logistic Regression 79.05 82.15 97.75 92.77 Gaussian Bayes Model 84 84 84 82 Random Forest 78 75 78 75 Decision Tree 80 73 75 73 Danish Logistic Regression 76.53 80.38 97.13 90.48 Gaussian Bayes Model 80.81 81 81 78 Random Forest 75 72 75 72 Decision Tree 76 70 72 70
 Accuracy (%) Precision (%) Sensitivity (%) Specificity (%) Cambridge Logistic Regression 84.42 87.36 97.70 55.70 Gaussian Bayes Model 77.4 74 77 70 Random Forest 86.5 90 86 87 Decision Tree 65.5 74 64 66 France Logistic Regression 87.58 90.36 90.35 82.55 Gaussian Bayes Model 67.50 66 67 61 Random Forest 80 87 80 80 Decision Tree 64 65.8 65.3 62 UK 84.52 87.49 93.07 56.62 Logistic Regression 84.52 87.49 93.07 56.62 Gaussian Bayes Model 77.45 75 77 70 Random Forest 86.5 90 86 87 Decision Tree 76 67 76 68 Danish Logistic Regression 82.69 85.76 91.98 55.83 Gaussian Bayes Model 75.3 73 75 68 Random Forest 84.5 89 84 85 Decision Tree 74 65 74 64

TABLE 6.13 Combination 3 (Without Considering WC)

 Accuracy (%) Precision (%) Sensitivity (%) Specificity (%) Cambridge Logistic Regression 82.09 84.84 97.55 93.27 Gaussian Bayes Model 83.16 84 83 80 Random Forest 87 86 86 86 Decision Tree 77 72 77 71 France Logistic Regression 87.58 90.36 90.35 82.55 Gaussian Bayes Model 71.40 73 71 72 Random Forest 69 70 69 69 Decision Tree 70 75 70 71 UK Logistic Regression 82.01 85.00 92.88 86.32 Gaussian Bayes Model 81.90 83 82 78 Random Forest 88 86 86 86 Decision Tree 76 71 76 71 Danish Logistic Regression 81.46 84.53 91.86 86.54 Gaussian Bayes Model 80.05 80 80 76 Random Forest 90 89 89 89 Decision Tree 75 71 75 71

Relation between different features for the Cambridge system is shown in Figure 6.11. The graph is plotted according to the pah using the correlation feature, as shown in Figure 6.12. A similar approach is applied to all the remaining T2D systems of European countries and then analyzed.

FIGURE 6.11 Correlation between the features for the Cambridge system.

The ROC curve is drawn for European countries with different algorithms, which is fully used to identify the risk score, as demonstrated in Figure 6.13.

The Cambridge system is cross-validated using AUC and its value is 0.8829. The France system is cross-validated using AUC and its value is 0.9408. The UK system is cross-validated using AUC and its value is 0.8962. The Danish system is cross-validated using AUC and its value is 0.8848.

Similarly, about 514,384 samples are used for analysis. Optimization of diabetes data using KNN classifier, DT, RF, SVM, and GB model is compared, as shown in Table 6.14.

FIGURE 6.12 Pair plot according to correlation feature values.

FIGURE 6.13 ROC curves showing the performance of the diabetes risk score in predicting diabetes.

 Machine Learning Algorithm Accuracy (Cambridge) Accuracy (France) Accuracy (UK) Accuracy (Danish) KNN classifier Training set: 0.99 Test set: 0.98 Training set: 1.00 Test set:1.00 Training set: 0.99 Test set: 0.99 Training set: 1.00 Test set: 0.99 Decision tr ee Training set: 1.00 Test set: 1.00 Training set: 1.00 Test set: 1.00 Training set: 1.00 Test set: 1.00 Training set: 1.00 Test set: 1.00 Random forest Training set: 1.00 Test set: 1.00 Training set: 1.00 Test set:1.00 Training set: 0.87 Test set: 0.86 Training set: 0.827 Test set: 0.827 Support vector machine Training set: 0.87 Test set: 0.86 Training set:0.875 Test set: 0.876 Training set: 0.965 Test set: 0.965 Training set: 0.95 Test set: 0.95 Gaussian Bayes model Training set: 0.7744 Test set: 0.7738 Training set: 0.6760 Test set: 0.6745 Training set: 0.7745 Test set: 0.7748 Training set: 0.7526 Test set: 0.7534

By considering the simple five parameters, namely, age, family history of diabetes, WC, physical activity, and BMI, the system was developed. Quickly, the data for these factors were obtained by five inquiries and scores acquired for these elements, as shown in Table 6.15.

TABLE 6.15 Diabetes Risk Score for European Countries

 Parameters Risk Score Age <35 0 35—49 15 >50 23 Obesity Waist Circumference Female <80 cm, Male <90 0 cm 17 Female 80-89 cm. Male 90-99 cm 20 Female >90 cm, Male >100 cm Physical Activity Vigorous Exercise 0 Mild Exercise 8 No Exercise 12 Family History Two Nondiabetic Parents 0 Either Parent Having Diabetes 19 Both Parents Having Diabetes 26 ВШ <25 0 25-29 11 30-34 15 >35 19 Maximum Score 100

Score >70:Very High Risk. 51-69: High Risk, 30-50: Medium Risk, <30: Low Risk

From this work, it is observed that performance of the system is better using logistic regression compared to other machine learning algorithms for both Asian and European countries. The corresponding bar chart of accuracy calculation using logistic regression for Asian and European countries is shown in Figures 6.14 and 6.15. For India, China, Sri Lanka, and Oman, the accuracy is 97.78%, 96.98%, 97.08%, and 97.175%, respectively. Similarly, for Cambridge, France, UK, and Danish, the accuracy is 84.42%, 87.58%, 84.52%, and 82.69%, respectively.

FIGURE 6.14 Accuracy using logistic regr ession for Asian countries.

FIGURE 6.15 Accuracy using logistic regression for European countries.

# CONCLUSION AND SCOPE OF FUTURE WORK

A simple diabetes risk assessment tool is developed and validated. A diabetes risk score system is developed for Asian and European countries using five different parameters, namely, age, WC, family history, physical activity, and BMI.

Several risk score tools are developed, but predicting the correct risk score without losing simplicity is really a challenging task. The proposed system is compared with the existing risk score system for the accuracy and performance. It can also be applied to the different ethnic groups. From this work, it is observed that performance of the system is better using logistic regression compared to other machine learning algorithms. In conclusion, the diabetes risk score system is developed that can be used in a stepwise screening strategy for T2D to provide individual age-specific personalized T2D risk score, using p coefficient for each year instead of making an age group. From Tables 6.9 and 6.15, it is concluded that score with <35 for Asian countries and <30 for European countries is considered as low risk, 35-69 for Asian countries and 30-50 for European countries is considered as medium risk, 70-95 for Asian countries and 51-69 for European countries is considered as high risk, and finally >95 for Asian countries and >70 for European countries is considered as very high risk. Optimum values >95 for Asian countries and >80 for European countries are identified using the ROC, and these values are validated using AUCs. This tool also provides the information about which factor affects T2D more. In future, the precautionary measure for this T2D will be provided by the expert panel.

# KEYWORDS

• type 2 diabetes
• risk score
• artificial intelligence

# REFERENCES

1. Л'. Mohan, R. Deepa. M. Deepa, S. Somannavar, and M. Datta, "A simplified Indian Diabetes Risk Score for screening for undiagnosed diabetic subjects,” J. Assoc. Phys. India, vol. 53, pp. 759-763, Sep. 2005.

• 2. J. Lindstrom and J. Tuomilehto, “The diabetes risk score: A practical tool to predict type 2 diabetes risk,” Diabetes Care, vol. 26, no. 3, pp. 725-731, Mar. 2003.
• 3. P. Katulanda, N. R. Hill, I. Stratton, R. Sheriff, S. D. N. De Silva, and D. R. Matthews, “Development and validation of a Diabetes Risk Score for screening undiagnosed diabetes in Sri Lanka (SLDRISK),” BMC Endocr. Disord.,vol. 16, no. 1, Jul. 25, 2016, Art. no. 42.
• 4. J. Al-Lawati and J. Tuomilehto, “Diabetes risk score in Oman: A tool to identify prevalent type 2 diabetes among Arabs of the Middle East,” Diabetes Res. Clin. Pract., vol. 77, no. 3, pp. 438-444, 2007.
• 5. S. J. Griffin, P. S. Little, C. N. Hales, A. L. Kinmonth, andN. J. Wareham, “Diabetes risk score: Towards earlier detection of type 2 diabetes in general practice.” Diabetes Metab. Res. Rev., vol. 16, no. 3, pp. 164-171, 2000.
• 6. H. Zhou, Y. Li. X. Liu, F. Xu, L. Li, K. Yang, X. Qian. R. Liu. R. Bie, and C. Wang, “Development and evaluation of a risk score for type 2 diabetes mellitus among middle- aged Chinese rural population based on the RuralDiab Study,” Sci. Rep., vol. 7, Feb. 17, 2017, Ait. no. 42685.
• 7. H. Bang. A. M. Edwards, A. S. Bomback, С. M. Ballautyne, D. Brillon. M. A. Callahan, S. M. Teutsch. A. I. Mushlin, and L. M. Kem. “A patient self- assessment diabetes screening score: Development, validation, and comparison to other diabetes risk assessment scores,” Ann. Internal Med., vol. 151. no. 11, pp. 775-783, 2009.
• 8. M. Rigla, G. Garcia-Saez, B. Pons, and M. E. Hernando. “Artificial intelligence methodologies and their application to diabetes,” J. Diabetes Sci. Techno!., vol. 12, no. 2, pp. 303-310, 2018.
• 9. L. Chen, D. J. Magliano, B. Balkau, S. Colagiuri. P. Z. Zimmet. A. M. Tonkin, P. Mitchell, P. J. Phillips, and J. E. Shaw, “AUSDRISK: An Australian Type 2 diabetes risk assessment tool based on demographic, lifestyle and simple anthropometric measures,” Med. J. Aust., vol. 192, no. 4, pp. 197-202, 2010.
• 10. N. Unwin, D. Whiting, L. Guariguata, G. Ghyoot, and D. Gan. Diabetes Atlas, 5th ed. Brussels, Belgium: International Diabetes Federation, 2011.
• 11. L. J. Gray, N. A. Taub, K. Khunti. E. Gardiner, S. Hiles. D. R. Webb, В. T. Srinivasan, and M. J. Davies, “The Leicester risk assessment score for detecting undiagnosed Type 2 diabetes and unpaired glucose regulation for use in a multiethnic UK setting,” Diabet Med., vol. 27, no. 8, pp. 887-895, 2010.
• 12. C. Gltimer, B. Carstensen, A. Saudbaek, T. Lauritzen. T. Jorgensen, and K. Borch- Johuseu, “A Danish diabetes risk score for targeted screening: The Inter99 study,” Diabetes Care, vol. 27, no. 3, pp. 727-733, 2004.
• 13. [Online.] Available: https://www. medicinenet. com type_2_diabetes/article. htm#what_ medicatious_treat_type_2_ diabetes
• 14. A. G. P. de Sousa. A. C. Pereira, G. F. Marquezine, R. M. do Nascimento-Neto, S. N. Freitas, R. L. de C. Nicolato, G. L. L. Machado-Coelho, S. L. Rodrigues, J. G. Mill, and J. E. Krieger,” Derivation and external validation of a simple prediction model for the diagnosis of type 2 diabetes mellitus in the Brazilian urban population,” Eur. J. Epidemiol, vol. 24, no. 2, pp. 101-109, 2009.
• 15. P. Adhikari, R. Pathak, and S. Kotian, “Validation of the MDRF—Indian Diabetes Risk Score (IDRS) in another south Indian population through the Boloor Diabetes Study (BDS),” J. Assoc. Physicians India, vol. 58, pp. 434-436, 2010.

16. Y. Y. Shanbhogue, S. Yidyasagar. M. Madken. M. Varma, С. K. Prashant, P. Seth, and

K. S. Natraj, “Indian Diabetic Risk Score and its utility in steroid induced diabetes,” J. Assoc. Physicians India, vol. 58, 2010, Ait. no. 202.

• 17. К. M. Sharma, H. Raujaui. H. Nguyen, S. Shetty, M. Datta, and К. M. Narayan, “Indian Diabetes Risk Score helps to distinguish type 2 from non-type 2 Diabetes Mellitus (GDRC-3),” J. Diabetes Sci. Technol., vol. 5. pp. 419-425, 2011.
• 18. M. J. Kim, N. K. Lim, S. J. Choi, and H. Y. Park. “Hypertension is an independent risk factor for type 2 diabetes: The Korean genome and epidemiology study,” Hypertens Res., vol. 38, pp. 783-789, 2015.
• 19. A. S. Shera, F. Jawad, and A. Maqsood, “Prevalence of diabetes in Pakistan.” Diabetes Res. Clin. Pract., vol. 76, no. 2, pp. 219-222, 2007.
• 20. M. I. Schmidt, В. B. Duncan, H. Bang, et al„ “Identifying individuals at high risk for diabetes: The atherosclerosis risk in communities study,” Diabetes Care, vol. 28. pp. 2013-2018,2005.
• 21. I. Contreras and J. Yehi, “Artificial intelligence for diabetes management and decision support,” J. Med. Internet Res., vol. 20, no. 5, Art. no. el0775, 2018.
• 22. J. J. Nalluri, D. Barh, V. Azevedo, and P. Ghosh, "miRsig: A consensus- based network inference methodology to identify pan-cancer miRNA-miRNA interaction signatures,” Sci. Rep., vol. 7, 2017, Ait. no. 39684.
• 23. N. Razavian, S. Blecker, A. M. Schmidt, A. Smith-McLallen. S. Nigam. and D. Sontag. “Population-Level prediction of type 2 diabetes from claims data and analysis of risk factors,” Big Data, vol. 3, no. 4, 2015.