Keywords

1 Introduction

An Electronic Health Record (EHR) holds the digital version of the patients’ medical paper report. They are real-time and instantly available to the users in a secured manner. It contains the medical and treatment histories of the patient right from the entry. This information includes demographics of the patient, medical history, medical diagnosis, medications, treatments, radiology images and test results of disease [1]. The main feature of the EHR is that it can track the medical history of the patient and can be shared among different medical organizations to yield a better diagnosis and treatment of the disease. The data stored in the EHR enable the practitioners to take better decisions on the disease. The Electronic Health Record (EHR) improves the treatment of the patient by reducing medical errors, duplicate treatment approaches, and delay in treatments. With the data stored in the database, a medical practitioner can apply machine learning algorithms to make effective decisions on treatments and medications. Classification approaches intelligently work to provide many accurate recommendations for short-term disease risks. Several reasons have made the authors work on an ensemble classifier to predict disease. Recent work in [2] has proposed an ensemble approach for predicting diabetes mellitus using Random Forest classifiers optimized with a genetic algorithm. Accurate recommendations are needed in the healthcare industry to provide a way better living to the people. Machine learning classifiers can be used in the analysis of predicting lung cancer among patients [3]. Machine learning algorithms with ensemble approaches have yielded good accuracy with success [4]. This paper totally focuses on providing a recommendation of drugs required by the patient to be taken for appropriate medication.

The proposed research is organized in the following sections. Section 2 discussed the related work in machine learning and recommendation systems, Sect. 3 describes the proposed methodology for drug recommendation, Sect. 4 describes the experimental evaluation and findings and finally concluded in Sect. 5.

2 Related Work

This section briefly discusses the recent works of machine learning and data mining approaches in health care.

Recent advances in biotechnology have led to the production of genetic data and medical information which are generated from the Electronic Health Records (EHR). Such data can be converted into valuable information using machine learning and data mining approaches. Kavakiotis et al. [5] investigated the applications of machine learning and data mining approaches for predicting, detecting, complications, and management. In the analysis, Support Vector Machine (SVM) outperformed in predicting and detecting diabetes complications. A technique for predicting the health of a fetal based on maternal medical history using machine learning approaches was presented in [6], the work comprised of gathering dataset from 96 pregnant women and applying the data to the machine learning classifiers. In the predictive analysis Decision Tree achieved higher accuracy than the other supervised and unsupervised classifiers under clinical analysis. Similar work is done in [7], where a health monitoring system is established to monitor the health of pregnant women to avoid future complications in health. The proposed hybrid machine learning technique established relationships between the physical and mental factors to prevent health complications in women during pregnancies. Kuteesa et al. [8] surveyed machine learning techniques in HIV clinical research; the study concluded that there is a steady increase in the production of genetic data which needs machine learning techniques for translating information to knowledge. Kaur [9] proposed a novel and smart healthcare framework using machine learning with advanced security features such as monitoring daily activity, access control, dynamic data encryption standards and validation at endpoints, the proposed was more effective in diagnosing disease. A machine learning technique for detecting type-2 diabetes was presented in [10], where a data-informed framework is proposed using machine learning via feature engineering, the proposed was found effective in identifying medical subjects from an Electronic Health Record (EHR). Darabi et al. [11] applied supervised machine learning algorithms to estimate the risk mortality of patients admitted in Intensive Care Units (ICUs), the model outperformed in predicting the mortality of patients admitted in ICUs. A recommendation system involves presenting the appropriate data or information to the user using data analysis, such work is represented in [12], where a recommendation system was proposed for recommending most appropriate news to the user using a novel news utility model, the proposed model outperformed the existing models in recommending news to the user. Guan [13] proposed a recommendation model using multiple view information for addressing content heterogeneity; the proposed model outperformed using ample experiments and visuals demonstration. A comparison of machine learning tree classifiers for diabetic mellitus is proposed in [14], where the classifiers such as Random Forest, C4.5, Random Tree, REP tree and Logistic Model Tree accuracies were compared in prediction. A multi-model technique for predicting chronic kidney disease was proposed by Kumar [15], which involves two-stage, such as genetic algorithm and machine learning for its disease prediction. A hybrid machine learning classifier was proposed in [16], which involves a multi-swarm optimization and a multi-layer perceptron for the prediction of DENV serotypes.

3 Proposed Methodology

We worked on a sample EHR dataset created only for testing purposes and not obtained from any source (medical practitioner), the dataset contained admission ID, Sex, Age, disease symptoms, and its prescriptions. First, the dataset is subjected to pandas tool for cleaning, which is then subjected to train various machine learning models for prediction, with the help of confusion matrix, the accuracy of the classifiers were calculated, and the classifier with high accuracy were applied to the test dataset for predicting and recommending drugs to the patients. The proposed framework is shown in Fig. 1.

Fig. 1
figure 1

Proposed framework

Full size image

The main steps in providing the recommendation for the disease are as follows:

  • Step 1: Importing and cleaning Dataset

In this step, the dataset was imported and cleaned using pandas, a python based data analysis tool.

  • Step 2: Training and visualizing of data.

In this step, the dataset is trained with its associated parameters and visualized using mathematical plotting functions to check the feasibility of data values.

A total of 891 testing records were visualized. Data visualizations help the data scientist to check the feasibility of the dataset attributes and its values. Figure 2 shows the relationship between males and female patients under analysis. The disease relationship between males and females is shown in Fig. 3. Symptom relationship between male and female is shown in Figs. 4 and 5, a prescription relationship between male and female is shown in Figs. 6 and 7. The scatter diagram for age and disease is shown in Fig. 8 and finally, the relationship between prescription and disease is shown in Fig. 9 respectively.

Fig. 2
figure 2

Male versus female

Full size image
Fig. 3
figure 3

Disease: male versus female

Full size image
Fig. 4
figure 4

Symptom 1: male versus female

Full size image
Fig. 5
figure 5

Symptom 2: male versus female

Full size image
Fig. 6
figure 6

Prescription 1: male versus female

Full size image
Fig. 7
figure 7

Prescription 2: male versus female

Full size image
Fig. 8
figure 8

Age versus disease

Full size image
Fig. 9
figure 9

Prescription versus disease

Full size image

  • Step 3: Applying supervised machine learning classifiers

In this step, the cleansed data are subjected to machine learning classifiers such as Support Vector Machine (SVM), Random Forest, Decision Tree, and K-nearest neighbor.

  • Step 4: Accuracy measurement

In this step, the classifiers were tested for its accuracy in recommending prescriptions for certain medical symptoms.

  • Step 5: Termination

Terminate the process when the machine learning classifier model is built for a recommendation.

figure a
figure b
figure c

The training algorithm is given in Algorithm 1 deals with training dataset which will then be given to the classifiers for generating appropriate drug recommendation, classifiers modeling algorithm in Algorithm 2 deals with producing classifier models for predicting disease, and finally, prediction and recommendation in Algorithm 3 deals with predicting disease and recommending suitable medication to the patient (Fig. 10).

Fig. 10
figure 10

Accuracy of the classifiers

Full size image

4 Experimental Analysis and Findings

This section contains the experimental results and analysis of a recommendation system for multi-disease. A dual-core i3 system with 4 GB of RAM, pandas, NumPy, Ipython, SciKit_Learn, SciPy, StatsModels, and Matplotlib libraries was used in the experimental analysis for generating recommendations. Machine learning classifier algorithms were evaluated for its accuracy in recommending appropriate drug recommendations to overcome certain medical issues. The experimental analysis takes place in two steps. In the first step, the dataset is imported and cleansed using pandas tool, the second step involves in training machine learning models for the recommendation system where the test data is subjected to various classifiers such as SVM, Random Forest, Decision Tree, and K-NN and accuracy is measured, then the classifier with high accuracy is used in predicting disease with various symptoms taken into consideration, with the trained symptoms and prescriptions, the Random Forest classifier provides recommendations to the patients (Table 1).

Table 1 Accuracy of classifiers
Full size table

5 Conclusion and Future Directions

In this work, a drug recommendation system for multi-disease using machine learning for healthcare was developed using a sample dataset which was created only for testing purposes and not obtained from any source (medical practitioner). The proposed method using Random Forest machine learning classifier showed that it can be an effective drug recommendation tool in healthcare. The machine learning classifiers used in the analysis include K-nearest neighbors, Support Vector Machine (SVM), Random Forest, and Decision Tree to achieve accuracy and provide drug recommendations for the patients suffering from short-term disease. The experimental results showed that the proposed method using the Random Forest classifier yielded a higher predictive performance compared with the other classifiers under analysis. Based on the experimental results obtained, the proposed method is found to be effective in improving the quality of drug recommendation, whereby improving the healthcare industry. In the future, we are interested in applying ensemble machine learning algorithms for predicting and generating drug prescriptions for multi-disease.