Article Preview
Top1. Introduction
The largest internal organ of our body is the liver which weights around 3 pounds (Liet al., 2012). Liver performs different types of metabolic functions like filtering blood, producing bile, assisting in fat digestion, making proteins for blood clotting, metabolizing drugs, storing glucose, and most importantly detoxifying harmful chemicals which were discussed by Singhetal.(2017).Due to malfunctioning of liver, it may cause liver disease and may affect our health seriously. Liver disease are caused due to various reasons like consumption of contaminated food, inherited disorders, accumulation of excessive fat, damaged hepatocytes which is infected with bacteria, virus or fungi, and consumption of alcohol or drugs in excess as discussed by Linet al.(2010). An early diagnosis of liver disease may increase the patient’s survival rate. To diagnose the liver disease along with various examination tests expert physicians are required, but it cannot assure the correct diagnosis was discussed by Takkaret. al (2017). Liver function tests majorly help in examining liver disorder. Key parameters in the test include albumin, alkaline phosphatase, total proteins, aspartate aminotransferase, alanine aminotransferase, direct bilirubin, total bilirubin, gamma-glutamyl transferase, prothrombin time, triglycerides platelet count and so on. Liver diseases are categorized into more than 100 type and the liver disease can be acute or chronic. Some liver disease has successful treatments while others don’t have. Liver disease is generally caused by accumulation of excess fat, consumption of alcohol on long term basis, consumption of contaminated food, inherited disorders, and drug overdose.
Intelligent systems play a significant role in medical industry in terms of disease detection and prediction. Data mining algorithms, neural networks, statistical techniques are widely implemented on liver examination data to evaluate the sickness. Predictive modeling has been one of the broadly used intelligent techniques for automated detection of multiple diseases like cancer, cardiac arrhythmia, liver disease, lungs infection etc. There is a need of making predictive models more accurate for disease predictions like chronic liver disease, cancer, lungs disease, heart disease etc. Machine learning calculations play a significant role to the specialists in providing essential measurements, continuous information, and progressed examination about the patients’ illness, lab test results, circulatory strain, clinical preliminary information, family history, and many more. Machine learning offers a guarantee for improving the detection and prediction of disease that has been made an interest in the biomedical field and also improve the decision-making process by increasing its objectivity. By way of machine learning techniques medical problems can be easily solved and thereby the diagnosis cost will be reduced. This study proposes an enhancement of the predictive models used in the original study by Ramana et al. (2011) to increase balanced accuracy and effectiveness of the models. The study focuses on two patient data sets (India and USA) and we have applied different machine learning techniques like Logistic Regression, K-Nearest Neighbors, Gradient Boosting Machine, Feedforward Neural Network, Support Vector Machine, C5.0, Naïve Bayes, Radio Frequency and the performance measures of these techniques were estimated on various perspectives such as Balanced Accuracy, Precision, recall, F1 – Score etc. to propose efficient classification algorithm for liver disease detection from various levels of enzymes, age and other factors and to predict the best classification algorithm in terms of accuracy, precision, specificity and sensitivity. Moreover, the performance was compared using the receiver operative characteristic (ROC). Classification algorithms are very effective and can be implemented in different automated medical diagnosis tools.
The aim of this study is to apply different machine learning techniques on liver datasets and thereby propose efficient classification algorithm. The remainder of this paper is structured in different sections where section 2 gives an idea of previous studies on liver disease diagnosis using classification algorithms. Section 3 describes the methodology and empirical design. Section 4 represents the finding of the study on liver datasets. Section 5 represents the implications of this study.