SRADHO: statistical reduction approach with deep hyper optimization for disease classification using artificial intelligence

Abstract

Artificial Intelligence techniques are being used to analyse vast amounts of medical data and assist in the accurate and early diagnosis of diseases. The common brain related diseases are faced by most of the people which affects the structure and function of the brain. Artificial neural networks have been extensively used for disease prediction and diagnosis due to their ability to learn complex patterns and relationships from large datasets. However, there are some problems like over-fitting, under-fitting, vanishing gradient and increased elapsed time occurred in the course of data analysis and prediction which results in performance degradation of the model. Therefore, a complex structure perception is much essential by avoiding over-fitting and under-fitting. This empirical study presents a statistical reduction approach along with deep hyper optimization (SRADHO) technique for better feature selection and disease classification with reduced elapsed time. Deep hyper optimization combines deep learning models with hyperparameter tuning to automatically identify the most relevant features, optimizing model accuracy and reducing dimensionality. SRADHO is used to calibrate the weight, bias and select the optimal number of hyperparameters in the hidden layer using Bayesian optimization approach. Bayesian optimization uses a probabilistic model to efficiently search the hyperparameter space, identifying configurations that maximize model performance while minimizing the number of evaluations. Three benchmark datasets and the classifier models logistic regression, decision tree, random forest, K-nearest neighbour, support vector machine and Naïve Bayes are used for experimentation. The proposed SRADHO algorithm achieves 98.2% of accuracy, 97.2% of precision rate, 98.3% of recall rate and 98.1% of F1-Score value with 0.3% of error rate. The execution time for SRADHO algorithm is 12 s.

Keywords: Bayesian optimization; Deep hyper optimization; Feature selection; Singular matrix; Statistical reduction approach.

Fig. 1

Methods for feature selection.

Fig. 2

Framework for feature selection and model evaluation.

Fig. 3

The overall working process of the proposed work.

Fig. 4

Finding the values of SRA using $m\times n$ matrix.

Fig. 5

(a) Identity matrix for $A$. (b) Identity matrix for $C$.

Fig. 6

Overview of statistical reduction approach.

Fig. 7

Statistical reduction approach using $A,BandC_{}^{\ast }$ matrices.

Fig. 8

Deep hyper optimization architecture in ANN.

Fig. 9

The true objective function.

Fig. 10

Samples from true objective function.

Fig. 11

Initiate the surrogate model.

Fig. 12

Maximize acquisition function to select the next point.

Fig. 13

Exploratory data analysis for Parkinson’s dataset.

Fig. 14

Exploratory data analysis for Alzheimer’s disease dataset.

Fig. 15

Exploratory data analysis for stroke prediction dataset.

Fig. 16

Statistical summary of the exploratory data analysis.

Fig. 17

Statistical analysis for age attribute and GSM119615 attribute.

Fig. 18

Comparison of accuracy using classifiers and SRADHO algorithm.

Fig. 19

Comparison of precision rate using classifiers and SRADHO algorithm.

Fig. 20

Comparison of recall rate using classifiers and SRADHO algorithm.

Fig. 21

Comparison of F1-Score using classifiers and SRADHO algorithm.

Fig. 22

False positive rate analysis for various datasets.

Fig. 23

False negative rate analysis for various datasets.

Fig. 24

True negative rate analysis for various datasets.

Fig. 25

Receiver operating characteristic curve for Parkinson’s dataset.

Fig. 26

Receiver operating characteristic curve for Alzheimer’s disease dataset.

Fig. 27

Receiver operating characteristic curve for Stroke Prediction dataset.

Fig. 28

Area under the curve values for each datasets.

Fig. 29

Analysis of Loss and Accuracy of Parkinson’s dataset using SRADHO.

Fig. 30

Analysis of loss and accuracy of Alzheimer’s disease dataset using SRADHO.

Fig. 31

Analysis of loss and accuracy of Stroke prediction dataset using SRADHO.

Fig. 32

Elapsed time for different datasets.