Patient care future: notes of textual mining secretions to reduce reading processes

Links table

Abstract and 1. Introduction

2. Related works

3. The methodology and 3.1 data

3.2 Preparation for data

3.3. Prediction models

4. Evaluation

4.1. Evaluation measures

4.2. Results and discussion

5. Conclusion and references

a summary

The hospital’s re -acceptance is the knowledge that they are patients with a re -hospital shortly after the exit, a decisive source of anxiety because it affects the patient’s results and health care costs. The identification of patients at risk of re -interventions allows for timely interventions, reduce hospitalization rates and total treatment costs. This study focuses on predicting the patient’s re -reading within less than 30 days using text mining techniques applied to the text report texts from electronic health records (EHR). Many methods of automated learning and deep learning were used to develop a classification model for this purpose. The new aspect of this research includes benefiting from the BIT (BDSS) vulva model as well as extracting the main component analysis feature (PCA) for pre -processing data for the introduction of a deep learning model. Our analysis of the MIMIC-II Data collection indicates that our approach, which combines the BDSS model with multi-layer cognitive (MLP), is outperforming modern methods. This model achieved 94 % summons and an area under the curve (AUC) by 75 %, which exposes its effectiveness in predicting the patient’s coercion. This study contributes to the progress of predictive modeling in health care by integrating text mining techniques with deep learning algorithms to improve the patient’s results and improve resource customization.

1. Introduction

The country’s health care sector is among the areas that impose significant costs on the government and insurance organizations annually [1]. The digitization of medical records for patients, in addition to enhancing the quality of medical services provided to citizens, will establish the basis for many cost savings in health care expenses in the country [2, 3]. Online healthcare services are highly dependent on EHR to store medical information for patients, exchange and manage them effectively [4]. One of the scales that were examined in the world of improving the quality of treatment and achieving financial savings is the use of electronic patient files to monitor the acceptance rate for patients in hospitals [5].

The hospital’s re -admission indicates the operation in which the patient returns, after being discharged from the hospital, and it is accepted in a short period and relatively shortened [6]. The hospital re -admission rate is currently an indication of the quality of the hospital’s performance in certain conditions. A high level of this scale has harmful effects on patient care costs [7]. Consequently, medical services centers have provided a plan to alleviate the hospital re -reading operations [8]. The goal of this approach is to enhance patient care quality and reduce medical care expenses. In cases where the patient’s re -reading rate exceeds acceptable levels, hospitals may face financial penalties [9, 10].

Various studies have searched in influential factors that contribute to the patient’s re -reading rate to hospitals and aims to predict the possibility of the patient’s acceptance within a short period after the exit (for example, 30 days, 70 days, and 90 days) [11-13]. Most of the predictive methods are primarily a visit to the patient’s last hospital. One of the effective methods of preventing the recurrent and repeated patients to the hospital is the use of artificial intelligence (AI) and its predictive approach. These methods allow prediction and identification of patients who are very likely to return to the hospital. After that, more accurate and appropriate therapeutic decisions can be made to reduce this possibility [14].

Text mining, as a sub -group of artificial intelligence, aims to address specific tasks such as identifying relevant documents and extracting important details from them, and contributing to efforts to solve problems within various fields [15]. In the field of health care, many studies have used text mining methods to classify patients into pre -specific groups or extract valuable visions from different sources such as clinical notes [16]Academic publications [17]And social media, where patients share their experiences and information [18].

In this study, using clinical observations, a set of machine learning models and deep learning models including logistical slope, random forests, K-Nearest (KNN), supporting support machine (SVM), naive gossip, and MLP to build a prediction model for re-acceptance. Our study offers several prominent contributions:

(1) A wide range of models: Unlike the previous approach, which mainly depends on machine learning, our research systematically compared a wide range of machine learning models and deep learning models. As a result, we have identified the most accurate alert model.

(2) Representative of the advanced textIn addition to the frequency of the range of the frequency document (TF-IDF), we used the BDSS model to include words to encrypt clinical text data. Moreover, we used PCA to reduce dimensions, which enhances mathematical efficiency.

(3) Data set size: I diverged from the previous methods, our study boldly published the proposed models through the entire data collection, even in the face of the great defect.

(4) Treating unbalanced data groups: In a blatant contradiction with the previous methodologies that lean on samples or excessive eating, our study has confidently from using data balance techniques. It is striking that our models showed a worthy performance, which confirms the durability and effectiveness of them.

(5) Input features: We used our achievement exclusively textual features, and avoiding any deliberate dependence on demographic features. This deliberate option allowed us to focus on the fundamental content of the data, which is not accumulated with external factors.

The rest of the paper is organized as follows: Section 2 provides a review of the relevant work. Section 3 shows the proposed form, which includes the introduction of the data set, detailing pre -processing steps, and describing the learning models used. Section 4 presents the results obtained from the experiments conducted and a discussion on the results. Finally, section 5 ends the paper.