Computational Intelligence Approaches to Predictive Modeling in Clinical Dataset Issues and Challenges: A Review

Shweta Kharya1, *, Sunita Soni1, Tripti Swarnkar2, Santosh Kumar Sar3, Sachi Nandan Mohanty4
1 Department of CSE, Bhilai Institute of Technology Durg, Chhattisgarh, India
2 Department of Computer Applications, Siksha' O' Anusandhan University, Bhubaneshwar, Odisha, India
3 Department of Engineering Chemistry, Bhilai Institute of Technology, Durg, India
4 Department of Computer Science, Singidunum University, Serbia

Abstract

Predictive modeling in clinical datasets presupposes the expansion of the computational system with the capability to analyze a massive amount of medical data to predict the outcome of every patient. Computational intelligence-based expert systems are acceptable for the analysis of complex data to process it fast and accurately as compared to conventional statistical methods. So the motivation to work with clinical datasets, characteristics like complexity, uncertainty, and imprecision, an advanced predictive model should be developed. This chapter provides a rigorous literature review of recent work on computational intelligence approaches applied to the clinical dataset using predictive modeling. Precisely this chapter's objective is concentrated only on those predictive computational intelligence approaches suitable for handling various characteristics and challenges in a clinical dataset like ever-changing data, fragmented data, interdependency among data, poor quality data, colossal volume and heterogeneity, and inaccessible data. Here exploration is done based on the prediction accuracy of a few computational intelligence approaches like Artificial Neural Networks, Deep Learning, Decision Trees, Support Vector Machines, Fuzzy based methods, as well as Bayesian approaches over many clinical datasets, especially breast cancer and its nature and suitability to work with clinical datasets are pointed out.

Keywords: Artificial Neural networks , Bayesian approaches , Breast cancer , Computational Intelligence , Clinical datasets , Decision Trees , Deep learning , Fuzzy techniques , Genetic algorithms , Predictive modeling , Support vector machines .

* Corresponding author Shweta Kharya: Department of CSE, Bhilai Institute of Technology Durg, Chhattisgarh, India; E-mail: shweta.bitdurg@gmail.com

INTRODUCTION

In today's digital world, data is all around but then also information lags and knowledge drops; so to do magic with this colossal data, there are various techniques and tools; among these is predictive modeling which is a ground-breaking way to leverage the hidden information to take an appropriate decision from a large number of data set. The colossal collections of electronic healthcare repositories are increasing the chances of developing an expert clinical decision support system that medical experts can use to enhance the patient's health care. The salient challenges for medical experts are diagnosing diseases, assessing risk, and evolving appropriate methods for prediction and final results. The goal of predictive modeling in health care necessitates advancing computational models with the strong capability of predicting future healthcare outcomes [1]. Statistical methods and computational intelligence can improve a new clinical decision support system using a new paradigm. The approaches of computational intelligence amalgamate metaheuristic optimization algorithms, such as Genetic algorithms and fuzzy logic, with machine learning algorithms, such as Artificial Neural Networks, Bayesian Models, and Deep Learning. Predictive modeling has a prime role in machine learning algorithms in which a model is built using existing datasets to make decisions on new patient data, such as a predictive model can be constructed to predict breast cancer using sets of input datasets with clinical results. Once the model is trained using the learning process and tested on standard datasets, the model is ready to receive new cases and predict clinical outcomes.

Computational intelligence approaches can tackle the potential to handle imprecision and uncertainty, which is entirely possible in the clinical dataset [2]. These approaches also work perfectly with a large and complex clinical dataset. Computational intelligence algorithms have been proposed to build predictive models, for example, prostate cancer [1], cardiovascular disease [3-5], lung cancer [6], diabetes [7, 8], and Alzheimer's disease [9].

This survey chapter discusses recent research on various clinical datasets like heart disease, chronic kidney disease, prostate cancer, lung cancer, and diabetes but rigorously reviewed papers on breast cancer predictive modeling using computational intelligence approaches. Breast cancer increases at an unpredictable rate, making it the most common and scary cancer in women [10, 11].

SIGNIFICANCE OF PREDICTIVE MODELING IN THE CLINICAL CARE INTELLIGENCE

Predictive modeling is the primary area of interest to all research communities and organizations. The massive availability of lots of new computational techniques and tools for predictive modeling assists researchers and practitioners in selecting the most appropriate strategy. Predictive analysis has become a noteworthy spectrum for researchers and practitioners in the clinical world. The following describes the significance of healthcare predictive analytics for healthcare providers, as shown in Fig. (1).

Fig. (1))
Benefits of predictive modeling in clinical care.

Improved Diagnostics

The predictive modeling system is beneficial in clinical decision-making. Few diseases have prototypical symptoms that can be easily identified and cured by qualified doctors according to the predefined treatment plan. But in some cases, patients have unconventional signs that point to a particular disease, making diagnostics more intricate. So, predictive modeling aims to derive new models for complex problems that can use lab testing details and diagnostic procedures to predict the outcome of interest. Thus, predictive analytics acquire a magnificent place in the treatment and diagnosis process.

Sky-High Price Effectiveness

Healthcare organizations implement a predictive model so that it can reduce costs significantly. Detailed information on cost management and patient risks can be generated, including a significant amount of accessible statistics on patients, employees, types of equipment, and planning.

Intense Operational Efficiency

When encumbered, hospitals have a medical staff paucity which affects the quality of the safe-keeping of patients. By developing expert models, the hospital can optimally allocate administrative assets and report to higher authorities about staffing provocation in advance. A software specialist can build a model by analyzing predictors in such cases.

Decreased Readmission Rates

The readmission rate indicates the standard services provided at a particular hospital. A hospital rewards fine to patients for patient readmissions in case of disease relapse in many countries according to the applicable regulations. Predictive models play a magnificent role in the organization while decreasing readmission rates by calculating the probability of readmission based on historical healthcare data.

Customized Clinical Care

The opportunities of precision medicine increase its efficiency in healthcare institutions. Based on personal health records, predictive modeling improves patient-centered care and contributes to creating the most effective nursing plans designed for every patient. Predictive prototypes are exceptionally efficient for inpatient and extremity treatment when there is a need for fast decisions.

Clinical Data Challenges

• Ever-changing Data- Clinical datasets are constantly updating, requiring methods to update the changes incrementally.
• Fragmented Data- As data are from various sources such as structured data, paper, digital, pictures, videos, and multimedia, extraction and integration of different data formats is a real challenge in the medical care sector.
• Interdependency among the Data- Correlation among different symptoms of clinical data exists highly [12].
• Faulty Data- Inaccurate data is most harmful to data analytics as mistaken input; the output is always untrustworthy. The leading cause of inaccurate is noise and missing data...