Advanced Machine Learning Methods Research Articles

An accessible and versatile deep learning-based sleep stage classifier.

Manual sleep scoring for research purposes and for the diagnosis of sleep disorders is labor-intensive and often varies significantly between scorers, which has motivated many attempts to design automatic sleep stage classifiers. With the recent introduction of large, publicly available hand-scored polysomnographic data, and concomitant advances in machine learning methods to solve complex classification problems with supervised learning, the problem has received new attention, and a number of new classifiers that provide excellent accuracy. Most of these however have non-trivial barriers to use. We introduce the Greifswald Sleep Stage Classifier (GSSC), which is free, open source, and can be relatively easily installed and used on any moderately powered computer. In addition, the GSSC has been trained to perform well on a large variety of electrode set-ups, allowing high performance sleep staging with portable systems. The GSSC can also be readily integrated into brain-computer interfaces for real-time inference. These innovations were achieved while simultaneously reaching a level of accuracy equal to, or exceeding, recent state of the art classifiers and human experts, making the GSSC an excellent choice for researchers in need of reliable, automatic sleep staging.

Comparison of Different Machine Learning Scenarios for Stock Price Prediction of Netflix

Buying stocks based on relatively accurate predictions to make a profit is what investors have been yearning for. However, due to the volatility and stochastic intrinsic of the stock, the price is also full of uncertainty, which is difficult to predict. With the improvement of computer performance and the popularization of machine learning methods nowadays, effective stock prediction methods emerge one after another. In this paper, random forest, XGBoost, and LSTM techniques are utilized for predicting the closing price of Netflix, which is one of the major stocks in Nasdaq and has been fluctuating this year. The closing price of the previous five days and the last day’s volume are used as inputs. The models are evaluated by MSE and R-square. According to the analysis, similar model evaluation results in random forest and LSTM indicate that the models are efficient. Furthermore, the prediction could be improved by considering the fluctuations of stocks simultaneously and introducing more variables in various dimensions. These results shed light on guiding further exploration of predicting stock price movements based on advanced machine learning methods.

Understanding Quality of Products from Customers’ Attitude Using Advanced Machine Learning Methods

The trend of E-commerce and online shopping is increasing rapidly. However, it is difficult to know about the quality of items from pictures and videos available on the online stores. Therefore, online stores and independent products reviews sites share user reviews about the products for the ease of buyers to find out the best quality products. The proposed work is about measuring and detecting product quality based on consumers’ attitude in product reviews. Predicting the quality of a product from customers’ reviews is a challenging and novel research area. Natural Language Processing and machine learning methods are popularly employed to identify product quality from customer reviews. Most of the existing research for the product review system has been done using traditional sentiment analysis and opinion mining. Going beyond the constraints of opinion and sentiment, such as a deeper description of the input text, is made possible by utilizing appraisal categories. The main focus of this study is exploiting the quality subcategory of the appraisal framework in order to predict the quality of the product. This paper presents a quality of product-based classification model (named QLeBERT) by combining quality of product-related lexicon, N-grams, Bidirectional Encoder Representations from Transformers (BERT), and Bidirectional Long Short Term Memory (BiLSTM). In the proposed model, the quality of the product-related lexicon, N-grams, and BERT are employed to generate vectors of words from part of the customers’ reviews. The main contribution of this work is the preparation of the quality of product-related lexicon dictionary based on an appraisal framework and automatically labelling the data accordingly before using them as the training data in the BiLSTM model. The proposed model is evaluated on an Amazon product reviews dataset. The proposed QLeBERT outperforms the existing state-of-the-art models by achieving an F1macro score of 0.91 in binary classification.

A survey on machine learning schemes for fiber nonlinearity mitigation in radio over fiber system

Abstract The fifth generation is the most recent generation of communication needed for high data rates. High spectrum availability, low jitter, high reliability, minimal latency, and increased capacity are just a few benefits of 5th generation. Optical fiber supports the 5G network’s backhaul to meet the enhanced capacity and big data rate requirements. When the fiber is used as a propagation medium, a significant number of nonlinearities manifest. These nonlinearity effects in optical fiber communication are among the most detrimental to modern communication systems because it results in various modulator distortions like phase, harmonic, and intermodulation, distortion, adjoining channel noise, and many more undesirable consequences. The primary fiber nonlinear effects in the radio over fiber communication systems are the Kerr nonlinearity and scattering effects, which are caused by alterations in refractive index as a result of the signal. To overcome these limits, several strategies have been put forth. In particular, the nonlinearity during signal modulation, transmission, and detection has attracted a lot of study attention due to the complex physical layer restrictions in RoF systems. One such fascinating potential is machine learning (ML) methods. In this article, we look at recent advances in ML methods for RoF systems, notably those that use ML models to reduce various types of impairments and improve system performance.

Unpacking the inter- and intra-urban differences of the association between health and exposure to heat and air quality in Australia using global and local machine learning models

Environmental stressors including high temperature and air pollution cause health problems. However, understanding how the combined exposure to heat and air pollution affects both physical and mental health remains insufficient due to the complexity of such effects mingling with human society, urban and natural environments. Our study roots in the Social Ecological Theory and employs a tri-environmental conceptual framework (i.e., across social, built and natural environment) to examine how the combined exposure to heat and air pollution affect self-reported physical and mental health via, for the first time, the fine-grained nationwide investigation in Australia and highlight how such effects vary across inter- and intra-urban areas. We conducted an ecological study to explore the importance of heat and air quality to physical and mental health by considering 48 tri-environmental confounders through the global and local random forest regression models, as advanced machine learning methods with the advantage of revealing the spatial heterogeneity of variables. Our key findings are threefold. First, the social and built environmental factors are important to physical and mental health in both urban and rural areas, and even more important than exposure to heat and air pollution. Second, the relationship between temperature and air quality and health follows a V-shape, reflecting people's different adaptation and tolerance to temperature and air quality. Third, the important roles that heat and air pollution play in physical and mental health are most obvious in the inner-city and near inner-city areas of the major capital cities, as well as in the industrial zones in peri-urban regions and in Darwin city with a low-latitude. We draw several policy implications to minimise the inter- and intra-urban differences in healthcare access and service distribution to populations with different sensitivity to heat and air quality across urban and rural areas. Our conceptual framework can also be applied to examine the relationship between other environmental problems and health outcomes in the era of a warming climate.

Influence analysis and relationship evolution between construction parameters and ground settlements induced by shield tunneling under soil-rock mixed-face conditions

This study investigates the influences of the key shield parameters on tunneling-induced ground settlements considering the dynamic excavation process. The advanced machine learning method in conjunction with principal component analysis is applied to evaluate the nonlinear relationships between ground settlements and the multi-parameters in the soil-rock composite ground. Considering the interaction among multi-factor, the nonlinear relationships of excavation characteristic parameters are developed by the correlation analysis to quantify shield machine performance in the mixed ground. Predictions on ground settlements in the soil-rock mixed-face are conducted in good agreement with the in-site measured data from the Nanjing Metro Line No. S7 tunnel. The implementation effects of the Light-GBM based model are evaluated by RMSE and R2. Subsequently, the influence weights of the most important influencing factors are ranked to control the parameters, which will be beneficial to optimize the excavation scheme during the dynamic construction process. Finally, the empirical formulas are proposed according to the high-correlation parameters as well as the measured datasets by the nonlinear multiple regression process, which can provide guidance for shield tunnel operation in the soil-rock composite ground.

Next-generation whole blood transcriptome-based assay for advanced adenoma detection.

77 Background: Colorectal Cancer (CRC) is the second leading cause of cancer mortality worldwide although highly curable if detected early. An accurate liquid biopsy test for early cancer and precancerous lesion, detection is key to promote prevention and reduce mortality. We previously developed a liquid biopsy test based on immune cells response to the tumor, commercialised under the name of Colox, with best-in-class clinical performances for AA detection (52% sensitivity and 92% specificity) (Ciarloni L et al, Clin Cancer Res, 2016). An improved second generation of the test, leveraging transcriptome profiling and advanced machine learning tools, is under development and first results are presented. Methods: Prospective peripheral whole-blood (PAXgene) samples from subjects diagnosed with CRC, advanced adenoma (AA), non-advanced adenoma, other types of cancer as well as controls without colorectal neoplasia (CON) were divided into discovery and validation sets. Transcriptome profiles were generated by RNA-seq and gene expression signatures identified leveraging Novigenix’s proprietary LITOseek platform, which integrates several advanced Machine Learning (ML) methods into a ranking systems. Biological functional analysis was performed by over-representation (ORA), gene set enrichment (GSEA) and gene network analyses (STRINGdb). Results: Significant changes in the whole blood transcriptome profile of AA compared to CON were identified. Functional analysis highlighted upregulation of fatty acid derivative biosynthesis, interferon signaling, tryptophan catabolism to kynurenine, and downregulation of DNA repair in AA blood samples compared to CON. A novel gene classifier was generated, showing for AA detection 71% sensitivity, 94% specificity and an AUC of 74% by cross-validation on the discovery set. Validation of the gene classifier in an independent set is currently ongoing and will be presented at the congress. Conclusions: Capturing immune-related information using whole blood trascriptome profiling and applying cutting-edge machine learning technologies demonstrated to be valuable for identifying gene signatures for advanced adenoma detection. This differentiated solution has demonstrated best-in-class potential to significantly improve patient outcomes by detecting precancerous lesions with a simple non-invasive blood test.

Integration of patient experience factors improves readmission prediction.

Many readmission prediction models have marginal accuracy and are based on clinical and demographic data that exclude patient response data. The objective of this study was to evaluate the accuracy of a 30-day hospital readmission prediction model that incorporates patient response data capturing the patient experience. This was a prospective cohort study of 30-day hospital readmissions. A logistic regression model to predict readmission risk was created using patient responses obtained during interviewer-administered questionnaires as well as demographic and clinical data. Participants (N = 846) were admitted to 2 inpatient adult medicine units at Massachusetts General Hospital from 2012 to 2016. The primary outcome was the accuracy (measured by receiver operating characteristic) of a 30-day readmission risk prediction model. Secondary analyses included a readmission-focused factor analysis of individual versus collective patient experience questions. Of 1754 eligible participants, 846 (48%) were enrolled and 201 (23.8%) had a 30-day readmission. Demographic factors had an accuracy of 0.56 (confidence interval [CI], 0.50-0.62), clinical disease factors had an accuracy of 0.59 (CI, 0.54-0.65), and the patient experience factors had an accuracy of 0.60 (CI, 0.56-0.64). Taken together, their combined accuracy of receiver operating characteristic = 0.78 (CI, 0.74-0.82) was significantly more accurate than these factors were individually. The individual accuracy of patient experience, demographic, and clinical data was relatively poor and consistent with other risk prediction models. The combination of the 3 types of data significantly improved the ability to predict 30-day readmissions. This study suggests that more accurate 30-day readmission risk prediction models can be generated by including information about the patient experience.

Interpretable time series forecasting of NPP parameters in accident scenarios

Nuclear power plant operators are limited in their capacity to analyze large amount of data in a short period, during a transient or accident scenario. Due to recent advances in machine learning methods, advanced algorithms have become available, which can support the operator with real-time analysis and forecast of important NPP parameters. The aim of this study was to provide short-term forecasts of key NPP parameters, which operators use to diagnose reactor states. The study focused on forecasting pressurizer level and subcooling margin, measured between the pressurizer and RCS hot leg, by using a state-of-the-art supervised learning algorithm, the Temporal Fusion Transformer. The model was trained with data from various transient and accident scenarios. The data was generated using an efficient basic principles PWR simulator called the BasicSim-PWR. The models demonstrated the capacity to differentiate between the small break LOCA and SGTR by predicting the safety trip of the SGTR scenario. The interpretable nature of Temporal Fusion Transformer gave insight into model behavior and potentially useful correlations for managing accident scenarios by observing few key inputs. The future research is directed towards advancing the scenario identification capability and efficiency evaluation of operator actions by forecasting the power plant response.

Wild Fires and Climate Change - Nowcasting and Forecasting Climate Change using Advances in Machine Learning Methods

Wild Fires and Climate Change - Nowcasting and Forecasting Climate Change using Advances in Machine Learning Methods

Network Intrusion Detection Model Using Fused Machine Learning Technique

With the progress of advanced technology in the industrial revolution encompassing the Internet of Things (IoT) and cloud computing, cyberattacks have been increasing rapidly on a large scale. The rapid expansion of IoT and networks in many forms generates massive volumes of data, which are vulnerable to security risks. As a result, cyberattacks have become a prevalent and danger to society, including its infrastructures, economy, and citizens’ privacy, and pose a national security risk worldwide. Therefore, cyber security has become an increasingly important issue across all levels and sectors. Continuous progress is being made in developing more sophisticated and efficient intrusion detection and defensive methods. As the scale of complexity of the cyber-universe is increasing, advanced machine learning methods are the most appropriate solutions for predicting cyber threats. In this study, a fused machine learning-based intelligent model is proposed to detect intrusion in the early stage and thus secure networks from harmful attacks. Simulation results confirm the effectiveness of the proposed intrusion detection model, with 0.909 accuracy and a miss rate of 0.091.

Non-Linear Associations Between the Urban Built Environment and Commuting Modal Split: A Random Forest Approach and SHAP Evaluation

The study of commuting mode choice is crucial since driving, with all its associated environmental and economic consequences, is the United States’ most popular mode of transportation due to urban sprawl, priority to road construction and America’s love affair with the automobile. More attention needs to be paid to sustainable modes such as public transit and walking. The built environment is expected to have an impact on commuting mode choice. Built environments with higher density, diversity, intentional design, destination accessibility, and shorter distance to transit (collectively known as the 5 Ds of the built environment) are hypothesized to lead to more sustainable mode choices, including public transit and walking. In this paper, we evaluate the impact of built environment variables on commuting modal split, including the four modes of public transit-bus, public transit-rail, walking, and driving. The study is conducted in Mecklenburg County, North Carolina, at the geographic level of census block groups in year 2015. Given the complexity of relationships in the built environment-travel behavior subject, the random forest method is used to predict aggregated commuting mode choice. Random forest is employed as it is capable of capturing nonlinear relationships and is not constrained by limitations in other widely used methods, such as multinomial logistic regression. After predicting the commuting mode shares, SHAP values (SHapley Additive exPlanations) are used to evaluate the impact of the built environment on commuting mode choices. As an advanced machine learning method, SHAP values adds explainability to the model. This method resolves the known limitation of machine learning methods as being “black boxes” and converts them to “white boxes” by providing interpretability. They provide insights into both the direction and magnitude of the relationships. Thanks to its rigorous ML-based design, our study helps to solidify the state of knowledge with strong evidence that block groups with higher degrees of the 5Ds lead to more choices of public transit and walking modes. We discuss urban policy implications of this study.

Postprocessing of Ensemble Weather Forecast Using Decision Tree–Based Probabilistic Forecasting Methods

Abstract Producing an accurate and calibrated probabilistic forecast has high social and economic value. Systematic errors or biases in the ensemble weather forecast can be corrected by postprocessing models whose development is an urgent challenge. Traditionally, the bias correction is done by employing linear regression models that estimate the conditional probability distribution of the forecast. Although this model framework works well, it is restricted to a prespecified model form that often relies on a limited set of predictors only. Most machine learning (ML) methods can tackle these problems with a point prediction, but only a few of them can be applied effectively in a probabilistic manner. The tree-based ML techniques, namely, natural gradient boosting (NGB), quantile random forests (QRF), and distributional regression forests (DRF), are used to adjust hourly 2-m temperature ensemble prediction at lead times of 1–10 days. The ensemble model output statistics (EMOS) and its boosting version are used as benchmark models. The model forecast is based on the European Centre for Medium-Range Weather Forecasts (ECMWF) for the Czech Republic domain. Two training periods 2015–18 and 2018 only were used to learn the models, and their prediction skill was evaluated in 2019. The results show that the QRF and NGB methods provide the best performance for 1–2-day forecasts, while the EMOS method outperforms other methods for 8–10-day forecasts. Key components to improving short-term forecasting are additional atmospheric/surface state predictors and the 4-yr training sample size. Significance Statement Machine learning methods have great potential and are beginning to be widely applied in meteorology in recent years. A new technique called natural gradient boosting (NGB) has been released and used in this paper to refine the probabilistic forecast of surface temperature. It was found that the NGB has better prediction skills than the traditional ensemble model output statistics in forecasting 1 and 2 days in advance. The NGB has similar prediction skills with lower computational demands compared to other advanced machine learning methods such as the quantile random forests. We showed a path to employ the NGB method in this task, which can be followed for refining other and more challenging meteorological variables such as wind speed or precipitation.

Clinical applications of artificial intelligence and machine learning in the modern cardiac intensive care unit

The depth and breadth of data produced in the modern cardiac intensive care unit (CICU) poses challenges to clinicians and researchers. Artificial intelligence (AI) and machine learning (ML) methodologies have been increasingly used to provide insights into this complex patient population. Major analytical tasks where ML methodology can be applied in the CICU and other critical care settings include mortality risk stratification, prognostication, non-fatal event prediction, diagnosis, phenotyping, identification of occult heart disease from the electrocardiogram and interpretation of echocardiographic images. In this review, we will discuss existing and future applications of different ML methods for CICU and other critical care populations, including penalized regression, standard ML methods (e.g., tree-based and other non-linear approaches) and advanced ML methods (e.g., deep learning and neural networks). While comparatively few published studies have applied ML methods in CICU populations, a more robust literature including patients with acute cardiovascular disease and non-cardiovascular critical illness can provide insights into CICU care. The CICU of the future is likely to utilize a sophisticated array of ML algorithms to streamline patient care by facilitating early recognition, diagnosis, phenotyping, and intervention for critically ill or deteriorating patients to improve providers’ cognitive load.

Credit-card Fraud Detection System using Neural Networks

Recently, with the development of online transactions, the credit-card transactions begun to be the most prevalent online payment methods. Credit-card fraud refers to the use fake Credit-Cards to purchase goods without paying. With the fast research and development in the area of information technology and data mining methods including the neural networks and decision trees, to advanced machine learning and deep learning methods, researchers have proposed a wide range of antifraud systems. Mainly, the Machine Learning (ML) and Deep Learning (DL) methods are employed to perform the fraud detection task. This paper aims to explore the existing credit-card fraud detection methods, and categorize them into two main categories. In addition, we investigated the deployment of neural network models with credit-card fraud detection problem, since we employed the Artificial Neural Network (ANN) and Convolutional Neural Network (CNN). ANN and CNN models are implemented and assessed using a credit-card dataset. The main contribution of this paper focuses on increasing the fraud-detection classification accuracy through developing an efficient deep neural network model.

Evaluating Social Determinants of Health Variables in Advanced Analytic and Artificial Intelligence Models for Cardiovascular Disease Risk and Outcomes: A Targeted Review

Introduction/PurposePredictive models incorporating relevant clinical and social features can provide meaningful insights into complex interrelated mechanisms of cardiovascular disease (CVD) risk and progression and the influence of environmental exposures on adverse outcomes. The purpose of this targeted review (2018–2019) was to examine the extent to which present-day advanced analytics, artificial intelligence, and machine learning models include relevant variables to address potential biases that inform care, treatment, resource allocation, and management of patients with CVD.MethodsPubMed literature was searched using the prespecified inclusion and exclusion criteria to identify and critically evaluate primary studies published in English that reported on predictive models for CVD, associated risks, progression, and outcomes in the general adult population in North America. Studies were then assessed for inclusion of relevant social variables in the model construction. Two independent reviewers screened articles for eligibility. Primary and secondary independent reviewers extracted information from each full-text article for analysis. Disagreements were resolved with a third reviewer and iterative screening rounds to establish consensus. Cohen's kappa was used to determine interrater reliability.ResultsThe review yielded 533 unique records where 35 met the inclusion criteria. Studies used advanced statistical and machine learning methods to predict CVD risk (10, 29%), mortality (19, 54%), survival (7, 20%), complication (10, 29%), disease progression (6, 17%), functional outcomes (4, 11%), and disposition (2, 6%). Most studies incorporated age (34, 97%), sex (34, 97%), comorbid conditions (32, 91%), and behavioral risk factor (28, 80%) variables. Race or ethnicity (23, 66%) and social variables, such as education (3, 9%) were less frequently observed.ConclusionsPredictive models should adjust for race and social predictor variables, where relevant, to improve model accuracy and to inform more equitable interventions and decision making.

Prediction of organic contaminant rejection by nanofiltration and reverse osmosis membranes using interpretable machine learning models.

Efficiency improvement in contaminant removal by nanofiltration (NF) and reverse osmosis (RO) membranes is a multidimensional process involving membrane material selection and experimental condition optimization. It is unrealistic to explore the contributions of diverse influencing factors to the removal rate by trial-and-error experimentation. However, the advanced machine learning (ML) method is a powerful tool to simulate this complex decision-making process. Here, 4 traditional learning algorithms (MLR, SVM, ANN, kNN) and 4 ensemble learning algorithms (RF, GBDT, XGBoost, LightGBM) were applied to predict the removal efficiency of contaminants. Results reported here demonstrate that ensemble models showed significantly better predictive performance than traditional models. More importantly, this study achieved a compelling tradeoff between accuracy and interpretability for ensemble models with an effective model interpretation approach, which revealed the mutual interaction mechanism between the membrane material, contaminants and experimental conditions in membrane separation. Additionally, feature selection was for the first time achieved based on the aforementioned model interpretation method to determine the most important variable influencing the contaminant removal rate. Ultimately, the four ensemble models retrained by the selected variables achieved distinguished prediction performance (R2adj = 92.4 %-99.5 %). MWCO (membrane molecular weight cut-off), McGowan volume of solute (V) and molecular weight (MW) of the compound were demonstrated to be the most important influencing factors in contaminant removal by the NF and RO processes. Overall, the proposed methods in this study can facilitate versatile complex decision-making processes in the environmental field, particularly in contaminant removal by advanced physicochemical separation processes.

SMART TEA: Churn, Trend, Inventory and Sales Prediction System Using Machine Learning

Managing operations at a tea factory requires consistency and planning. This paper presents a complete platform that uses advanced machine learning methods specifically designed for the tea sector. Sales prediction, churn prediction, trend prediction, and smart inventory management are the four essential features of our solution. While using Neural Networks for Churn Prediction offers exact insights into customer churn, utilizing Gradient Boosting for Sales Prediction guarantees accurate revenue estimates. Linear regression models were used for trend prediction and smart inventory management to enable efficient utilization of resources and trend identification. With the help of this integrated system, tea companies can now operate more profitably and sustainably in a market that is always changing. This research acts as a beacon, demonstrating the revolutionary potential of data-driven management as operations in the tea industry evolve.

Enhancing Predictive Modeling in Kuala Lumpur Real Estate: A Comprehensive Data Preprocessing and Feature Engineering Approach

The real estate market in Kuala Lumpur exhibits complex dynamics influenced by various factors. This paper presents a comprehensive approach to enhance the prediction of real estate prices in Kuala Lumpur utilizing a dataset of 49,416 records with advanced data preprocessing techniques, exploratory data analysis (EDA), and predictive modeling. Initially, the dataset undergoes rigorous preprocessing including handling missing values through property type-specific mean imputation, label encoding categorical variables, and standardization to ensure uniformity and compatibility for analysis. Subsequently, EDA techniques are employed to gain insights into the dataset's characteristics and relationships among variables. To improve model performance and interpretability, feature selection is performed based on Mutual Information (MI) score and correlation metrics. This aids in identifying the most relevant features for predicting real estate prices in Kuala Lumpur. Additionally, feature engineering techniques are applied to create two new features that capture nuanced aspects of the real estate market. The predictive modeling phase employs a Linear Regression algorithm to forecast real estate prices. Leveraging the preprocessed data and optimized feature set, the model aims to accurately predict property prices based on available features. The linear regression model offers interpretability, enabling stakeholders to understand the driving factors behind price variations. Through a rigorous methodology encompassing data preprocessing, exploratory analysis, feature selection, engineering, and predictive modeling, this study contributes to enhancing the accuracy and interpretability of real estate price prediction in Kuala Lumpurthat applies advanced machine learning methods. The findings offer valuable insights for real estate investors, policymakers, and stakeholders to make informed decisions in this dynamic market landscape.

Identification of Smoking-Associated Transcriptome Aberration in Blood with Machine Learning Methods.

Long-term cigarette smoking causes various human diseases, including respiratory disease, cancer, and gastrointestinal (GI) disorders. Alterations in gene expression and variable splicing processes induced by smoking are associated with the development of diseases. This study applied advanced machine learning methods to identify the isoforms with important roles in distinguishing smokers from former smokers based on the expression profile of isoforms from current and former smokers collected in one previous study. These isoforms were deemed as features, which were first analyzed by the Boruta to select features highly correlated with the target variables. Then, the selected features were evaluated by four feature ranking algorithms, resulting in four feature lists. The incremental feature selection method was applied to each list for obtaining the optimal feature subsets and building high-performance classification models. Furthermore, a series of classification rules were accessed by decision tree with the highest performance. Eventually, the rationality of the mined isoforms (features) and classification rules was verified by reviewing previous research. Features such as isoforms ENST00000464835 (expressed by LRRN3), ENST00000622663 (expressed by SASH1), and ENST00000284311 (expressed by GPR15), and pathways (cytotoxicity mediated by natural killer cell and cytokine-cytokine receptor interaction) revealed by the enrichment analysis, were highly relevant to smoking response, suggesting the robustness of our analysis pipeline.