Learning Techniques – Classification of Insulin-Dependent Diabetes Mellitus among adults
C.LALITHA 1 P. ANBUMANI 2
1 Assistant Professor, Department of Computer Application, Tagore College of Arts and Science, Chrompet, Chennai-600 044.
2 Research Scholar, Department of Computer Science, Periyar University, Salem – 636 011
Abstract: Insulin-dependent diabetes mellitus (IDDM) is a set of correlated diseases in which the body cannot regulate the amount of sugar in the blood. It mainly affects the adult and characterized by chronic hyperglycemia associated with disturbances of carbohydrate, fat, and protein metabolism due to absolute or relative deficiency in insulin secretion and/or action. It causes long term damage, dysfunction and failure of various organs such as eyes, kidneys, nerves, heart and blood vessels. A new methodology is used to find the stages of Insulin-dependent diabetes mellitus using Convolutional Neural Network (CNN). The symptoms and stages of Insulin-dependent diabetes mellitus are classified by using CNN technique. The hidden layers of a CNN typically consist of convolutional layers, pooling layers, fully connected layers and normalization layers. If you have fully-connected layers at the end of your convolutional network, implementing dropout is easy. It helps us to know the various stages of Insulin-dependent diabetes mellitus and to predict the recommend preclusion to patients those who are affected by Insulin-dependent diabetes mellitus and provide implication to that patient.
Index Terms — Convolutional neural network (CNN), Insulin-dependent diabetes mellitus (IDDM).
Insulin-dependent diabetes mellitus is a disorder caused when the body doesn’t make enough insulin. This high blood sugar produces the classical symptoms of polyuria (frequent urination), polydipsia (increased thirst) and polyphagia (increased hunger).If left untreated, diabetes can lead to blindness, kidney disease, nerve disease, heart disease, and stroke. Insulin-dependent diabetes mellitus often simply referred to as diabetes is a condition in which a person has high blood sugar, so that the body needs a special concern and provides extensive solutions to remain staying healthy, effervescent and lucid. This learning finds the knowledge by using supervised and unsupervised learning algorithms such as, Convolutional neural network. To find the efficiency of neural network technique by using the Root Mean square error and Mean absolute error 4 5. Beyond that the low level of attributes in Patient’s records falls in border phase of IDDM. Convolutional neural network has been employed to envisage the knowledge about the disease. Symptoms and stages are a class variables used for classification 19 20. A network composed of more than one layer of neurons, with some or all of the outputs of each layer connected to one or more of the inputs of another layer. The first layer is called the input layer, the last one is the output layer, and in between there may be one or more hidden layers.
IDDM – APPROACH
Figure 1: Overall architecture diagram
In Phase I, the patient’s dataset has been collected from diabetologists. The patient’s dataset contains the symptoms of IDDM such as Excessive thirst, Fatigue, hayfever, Dry Skin, Blurred vision, sudden weight loss, tingling hands and feet etc. The information has been examined and used in this experiment. In this dataset, demographic information is gathered from the patients. The result dataset contains the demographic and patient’s symptoms information is used for analysis 17 18.
In Phase II and III, Predictive and Descriptive learning methods are used to find the impact of disease in the urban and rural area. The methods are one of the Convolutional neural network respectively and it is allocated to IDDM stages of the patients 15. The six sigmoid nodes are used as inputs, the weights are assigned to each node, and their six output layers are classified based on IDDM stages. According to the stages based classification, the ten sigmoid nodes are used as input and three output layers are classified based on the stages of IDDM 12. In Phase IV, the association between symptoms and IDDM stage is analyzed using motivating measures.
II. REVIEW OF RELATED MODELING IMPETUS
2.1 Hidden layer activation function
To use a logistic (sigmoid) activation function for the hidden layers. A logistic function is recommended. Here Figure: 2 show the plot of a logistic activation function:
Figure 2: Logistic Activation Function
2.2 The Output Report Generated Using IDDM dataset
2.2.1 Project Parameters
============ Project Parameters ============
Target variable: STAGES OF IDDM
Number of predictor variables: 9
Type of model: convolutional Neural Network (CNN)
Number of layers: 3 (1 hidden)
Hidden layer activation function: Logistic
Output layer activation function: Logistic
Type of analysis: Classification
Category weights (priors): Data file distribution
Misclassification costs: Specified cost matrix
Validation method: Random sampling (90%)
Input data file: J:IDDM documentsDATASET.csv
Number of variables (data columns): 10
Data sub setting: Use all data rows
Number of data rows: 300
The Project Parameters section of this IDDM report displays a summary of the options and parameters user can selected on the various property pages for the model. It reveals that the classification techniques are employed for finding the predicted knowledge using CNN. In this model, Nine predictor variable are used and one target for variable for prediction. Random sampling method is used for validating the data. Only three layers are used for finding the knowledge from this modeling using logistic activation.
2.2.2 Summary of Variables
No. Variables Class Type Missing Rows Categories
1 Patients Predictor Categorical 0 30
2 Excessive thirst Predictor Continuous 0 12
3 Urinary Infection Predictor Continuous 0 2
4 Wrinkles Predictor Continuous 0 2
5 Skin problems Predictor Continuous 0 2
6 Fatigue Predictor Continuous 0 2
7 Hunger Predictor Continuous 0 2
8 Hayfever Predictor Continuous 0 2
9 yeast infections Predictor Continuous 0 2
10 Stages of IDDM Target Categorical 0 3
Table 1: Summary of Variables in IDDM datasets –Using DTREG
The Table 1 displays information about each variable in the IDDM dataset. The first column shows the name of the variable, the second column shows how the variable was used; the possibilities are Target, Predictor, Weight and Unused. The third column shows whether the variable is categorical or continuous, the forth column shows how many data rows had missing values on the variable, and the fifth column shows how many categories (discrete values) the variable has. In the case of continuous variables, the number of categories (such as Patients and Stages of IDDM) will be limited by the value specified for “Max. Categories for predictor variables” on the model design property page.
2.2.3 Classification Summary
It reveals that misclassification does not exist in the validation. The cost and weight information are tabulated based on classification in Table 2 and 3.
Category Actual category Misclassified category Percentage of cost
Count weight Count weight percent cost
Primary 99 99 0 0 0 0
Nonsevere 63 63 0 0 0 0
Severe 108 108 0 0 0 0
total 270 270 0 0 0 0
Table 2: Classification Table for Training Data
Category Actual category Misclassified category Percentage of cost
Count weight Count weight percent cost
Primary 11 11 0 0 0 0
Nonsevere 7 7 0 0 0 0
Severe 12 12 0 0 0 0
Total 30 30 0 0 0 0
Table 3: Classification Table for Validating Data
In this dataset, 30 objects are used for training and 270 objects are used for testing. Each category is classified and finds the weights are calculated using convolutional neural network.
2.2.4 Confusion Matrix Table
A “Confusion Matrix” provides detailed information about how data rows are classified by the model. The matrix has a row and column for each category of the target variable. The categories shown in the first column are the actual categories of the target variable. The categories shown across the top of the table are the predicted categories. The numbers in the cells are the weights of the data rows with the actual category of the row and the predicted category of the column. Here table 4 shows the IDDM datasets- confusion matrix. The numbers in the diagonal cells are the weights for the correctly classified cases where the actual category matches the predicted category. The off-diagonal cells have misclassified row weights. For IDDM dataset, the Non severe category was slightly misclassified as Primary and severe category.
Actual category Predicted Category
Primary Non severe Severe
Training Data Primary 11 0 0
Non Severe 0 7 0
Severe 0 0 12
Validation Data Primary 99 0 0
Non Severe 0 63 0
Severe 0 0 108
Table 4: Confusion Matrix
2.2.5 Variable Importance Table
The variable importance table gives a ranking of the overall importance of the predictor variables.
SL.NO VARIABLES IMPORTANCE
1 Excessive thirst 71.734
2 Urinary Infection 34.509
3 Wrinkles 33.083
4 Skin problems 33.083
5 Fatigue 30.584
6 Hunger 10.567
7 Hayfever 8.731
8 yeast infections 0.566
Table 5: Variables of Importance
Importance scores are computed by using information about how variables were used as primary splitters and also as surrogate splitters. If a primary splitter is slightly better than a surrogate, then the primary splitter may “mask” the significance of the other variable. By considering surrogate splits, the importance measure calculated by giving more accurate measure of the actual and potential value of a predictor 21. To get the most accurate measure of importance, the user should select the option “Always compute surrogate predictors” on the Missing Data property page. The importance score for the most important predictor is scaled as 71.734. Other predictors will have lower scores. Only predictors with scores greater than zero.
III. Results and Discussion
Based on the data set, the diagnosis was made by a physician with training and qualifications in diabetologists. For the purpose of this study, children and teenagers with IDDM and atopic dermatitis are excluded. Standard treatment is advised for the children and teenagers which is the same as children and teenagers with IDDM. The physician’s interpretation of clinical data and clinical images are stored in the medical databases22 23. An expert medical knowledge and specialized learning techniques to understand the meaning of unstructured data explanation. The dataset contains 300 patients’ objects. The technique has been employed to categorize the patients and their symptoms using predictive modeling software and it evaluates based on the errors occurred in the classification, which is shown in table 6.
SL.No Types of Error Value
1. Mean Absolute Error 0.0125
2. Root Mean Squared Error 0.0133
3. Relative Absolute Error 0.5362
4. Root Relative Squared Error 0.7089
Table 6: Error Calculation
Here the main notified errors are root mean squared and mean absolute error, which are minimum in this model. This reveals that the model classifies the dataset perfectly. The finding divulges that densely populated group of patients with close similarities based on the stages of IDDM are classified.
In this Paper, popular learning methods are used to predict the patient information. In Convolutional neural network technique, the three stages are identified as primary, non severe and severe8. It divulges that there is a perfect classification. Standard treatment is advised for the IDDM affected patients 3 especially for adults (;20). It comprises general recommendation on diet changes, regular exercise, possibly insulin shots and hormone tablet for acute flare-ups of the patients 14. The correlation study unveils that the symptoms and stages have strapping association. The association between each entity is identified by using classification 16.
1 Cardona, F., Morcillo, S., Gonzalo-Martin, M. and Tinahones, F. (2005). The apolipoprotein E genotype predicts postprandial hypertrigly ceridemia in patients with metabolic syndrome. Journal of Endocrinology and Metabolism, 90(5), 2972-2975
2 Dandona, P., Aljada, A., Chaudhuri, A., Mohanty, P. and Garg, R. (2005). Metabolic syndrome: A comprehensive perspective based on interactions between obesity, diabetes and inflammation. Circulation, 111(11), 1448-1454.
3 Homer J, Jones A, Seville D, Essien J, Milstein B, Murphy D. 2004. The CDC diabetes system modeling project: Developing a new tool for chronic disease prevention and control. 22nd International Conference of the System Dynamics Society, Oxford, England.
4 Ida J. Hatoum, Frank B. Hu, Jeanenne J. Nelson, and Eric B. Rimm, “Lipoprotein-Associated Phospholipase A2 Activity and Incident Coronary Heart Disease Among Men and Women With NIDDIDDMiabetes”, Diabetes VOL 59,May 2010.
5.C. Deeb, “Diabetes Technology During the Past 30 Years: A Lot of Changes and Mostly for the Better”, Diabetes L Spectrum, 2008. 21, pp 78- 83 (2008).
6 L. I. Kuncheva, C. J.Whitaker, C. A. Shipp, and R. P.W. Duin, “Is independence good for combining classifiers?,” in Proc. Int. Conf. Pattern Recognition (ICPR), vol. 2, Barcelona, Spain, 2001, pp. 168–171.
7 Miller, A., Blott, B., ; Hames, T.. “Review of Neural Network Applications in Medical Imaging and Signal Processing. Medical and Biological Engineering and Computing”, (1992)30(5), 449- 464.
8 Moriarty DG, Zack MM, Kobau R. 2003. The Centers for Disease Control and Prevention’s Healthy Days measures: Population tracking of perceived physical and mental health over time. Health and Quality of Life Outcomes 1(1): 37.
9 S-H.Min, I.Han: Optimizing Collaborative Filtering Recommender Systems Lecture Notes in Artificial Intelligence vol. 3528, 2005, pp.313–319.
10 Stewart WF, Ricci JA, Chee E, Hirsch AG, Brandenburg NA (June 2007). “Lost productive time and costs due to diabetes and diabetic neuropathic pain in the US workforce”. J. Occup. Environ. Med. 49(6):672–9. doi:10.1097/JOM.0b013e318065b83a. PMID 17563611.
11 Tang, W., Hong, Y., Province, M., Rich, S., Hopkins, P., Arnett, D., Pankow, J., Miller,
M. and Eckfeldt, J. (2006). Familial clustering for features of the metabolic syndrome. Diabetes Care, 29(3), 631-636.
12 Williams. D. Prevost, T., Whichelow, M., cox, B., Day, N. and Wareham, N. (2000). A
cross-sectional study of dietary patterns with glucose intolerance and other features of the metabolic syndrome Abstract. British Journal of Nutrition, 83(3), 257-266.
13Wilson, P., D’Agostino, R., Parise, H., Sullivan, L., and Meigs, J. (2005). Metabolic
Syndrome as a precursor of cardiovascular disease and NIDD IDDM diabetes mellitus. Circulation, 112, 3066-3072.
14 Adarsh, P., Jeyakumari, D.. Multiclass svm-based automated diagnosis of diabetic retinopathy. In: Communications and Signal Processing (ICCSP), 2013 International Conference on. IEEE; 2013, p. 206–210
15 Wong WL, Su X, Li X, Cheung CMG, Klein R, Cheng C-Y, et al. Global prevalence of age-related macular degeneration and disease burden projection for 2020 and 2040: a systematic review and meta-analysis. Lancet Glob Health. 2014; 2: e106–116. https://doi.org/10.1016/S2214-109X(13)70145-1 PMID: 25104651
16 Abràmoff MD, Lou Y, Erginay A, Clarida W, Amelon R, Folk JC, et al. Improved Automated Detection of Diabetic Retinopathy on a Publicly Available Dataset Through Integration of Deep Learning. Invest Ophthalmol Vis Sci. 2016; 57: 5200–5206. https://doi.org/10.1167/iovs.16-19964 PMID: 27701631
17 E. Renard, J. Place, M. Cantwell, H. Chevassus, and C. C. Palerm, “Closed-loop insulin delivery using a subcutaneous glucose sensor and intraperitoneal insulin delivery feasibility study testing a new model for the artificial pancreas,” Diabetes Care, vol. 33, no. 1, pp. 121–127, 2010
18 R. Hovorka, “Closed-loop insulin delivery: From bench to clinical practice,” Nature Rev. Endocrinol., vol. 7, no. 7, pp. 385–395, 2011.
19 A Sensitivity Analysis of (and Practitioners’ Guide to) Convolutional Neural Networks for Sentence Classification IJCNLP2017 Yingjie ZhangByron C. Wallace
20 Y. Kim, “Convolutional neural networks for sentence classification,” arXiv preprint arXiv:1408.5882, 2014.
21 A. Krizhevsky, I. Sutskever, and G. E. Hinton, “Imagenet classification with deep convolutional neural networks,” in Advances in neural information processing systems, 2012, pp. 1097–1105.
22 Akm Ashiquzzaman, Abdul Kawsar Tushar, Md Rashedul Islam, Dongkoo Shon, Kichang Im, Jeong-Ho Park, Dong-Sun Lim, Jongmyon KimReduction of overfitting in diabetes prediction using deep learning neural network. IT Convergence and Security, Springer (2018), pp. 35-43
23MichaelA. Pfeifer, Daniel Cook, Joel Brodsky, David Tice, A Reenan, Sally Swedine, JeffreyB.Halter, Daniel Porte”Quantitative evaluation of cardiac parasympathetic activity in normal and diabetic man Diabetes, 31 (4) (1982), pp. 339-345
About the Authors:
C.LALITHA is working as an Assistant Professor in the Department of Computer Application, Tagore College Arts and Science, Chrompet, Chennai. She has published many research articles in the National/International conferences and journals. Her research interests include Learning Techniques, Data Mining.
P. ANBUMANI is working as an Assistant Professor in the Department of Computer Application, Tagore College Arts and Science, Chrompet, Chennai. He has Pursuing Part-Time PhD in Department of Computer Science, Periyar University, Salem. He has published many research articles in the National conferences. Her research interests include Learning Techniques.