Application of Random Forest Model in Cancer Risk Prognosis: A Role for Dietary Habits

doi:10.21203/rs.3.rs-3906286/v1

Purpose- This study focuses on the impact of lifestyle behaviors, particularly eating habits, on cancer development. In recent years, there has been growing interest in predicting cancer risk using machine learning algorithms and analyzing factors such as diet.

Methods- Our research utilizes a Random Forest Model to classify and identify hidden risk factors in a sample of 252 individuals from the Semnan province in Iran, split into case and control groups.

Results- Seventeen dietary indicators were derived from nutritional questionnaires and used to diagnose cancer. The optimal number of trees in the model was 93, resulting in a model with high accuracy and predictive capability. The developed model achieved an average accuracy rate of 92% through cross-validation. Notable early predictors of cancer were discovered, including the extent of vegetable frying, methods of vegetable storage, and the types of containers used for daily staples like bread in Iranian cooking practices.

Conclusion- Based on these findings, it is important to address and implement targeted lifestyle interventions based on early predictors to improve patient well-being and treatment outcomes. This research highlights the potential gap in current recommendations regarding health and diet for cancer patients and emphasizes the need for tailored interventions.

carcinogens

food

dietary habits

classification

random forest model

Cancer, as one of the significant challenges in the global healthcare system, exhibits a rapidly growing incidence and a substantial contribution to the mortality rates of patients. In 2020, 19.3 million new cases and 10 million cancer-related deaths were reported worldwide[1, 2]. Global population aging, and the rising prevalence of key risk factors have led to an increasing burden of cancer incidence worldwide, particularly among Western Asian countries [3].

Iran located in the Middle East, faces one of the highest cancer growth rates in such a way that there are around 131,000 new cases registered annually, with nearly 79,000 individuals surrendering to the disease. Currently, 320,000 people are living with cancer [4] and based on GLOBOCAN estimates, there is an expected 17.3% increase in new cancer cases in Iran by the year 2025, under the assumption that current rates remain constant [5].

While genetics play an irrefutable role, scientific evidence highlights unhealthy lifestyles and high-risk behaviors as the primary cancer risk factors in more than 45% of malignancies [6]. These factors are intertwined with individuals' social and economic status, dietary habits, cultural beliefs, healthcare-seeking practices, alcohol and drug/tobacco use, and stress levels [7, 8]. Nutrition as an outstanding factor contributes to 30–40% of cancer types. In addition to the type of food consumed, dietary behaviors, including food preparation methods, calorie balance, dietary diversity, and portion sizes, can act as influential risk factors for cancer [9, 10].

‘Data mining’ has been increasingly applied in the medical literature recently. In general, data mining in biological research and evidence-based medicine uses big data analysis tools to predict various health outcomes by emphasizing variables that may not be identified using traditional statistical methods [11–13]. Within the spectrum of data mining techniques, the Random Forest model, serving as a machine learning algorithm, has demonstrated remarkable accuracy in discerning carcinogenic factors across various features. Its algorithms present a valuable tool for developing and performing efficacious policies aimed at cancer prevention within healthcare systems [14]. For instance, the Random Forest-based rule extraction method has been used for breast cancer diagnosis [15, 16]. Moreover, there are examples of data mining applications in cancer prediction with a focus on dietary factors. In a study by Zutphen et al [17] lifestyle behaviors associated with the recurrence and survival of colon cancer were identified by a Random Forest model. Nonetheless, insufficient data has analyzed behaviors associated with feeding and dietary habits, such as the degree of frying vegetables or the type of cookery and food containers, among individuals both with and without cancer.

Given Iran's extensive geographical range and its location on the cancer belt, and reported disparities by residence region in specific cancer burden in other studies (Chen et al., 2019), thorough regional surveys into the etiology of cancer are believed necessary. This study in Semnan province investigated feeding-related risk factors associated with cancer incidence, considering different dietary habits and carcinogenic exposures in Garmsar, Aradan, and Damghan cities.

Subjects

Participants in the present study, which was a case-control one, were 252 individuals living in Semnan province of Iran. A sample of 126 people with diagnosed cancer who were registered in the cancer registration system of the Iran Ministry of Health in the last three years were included in the case group (cancer), and 126 people from the same areas who were age and gender-matched were classified in the control group (without cancer).

Data gathering

Self-reported questionnaires were used to collect demographic, behavioral, and nutritional information. Nutritional data was gathered using a researcher-made questionnaire (used in the pilot study on the case group, and its validity and reliability were confirmed with a Cronbach's coefficient of more than 70%). Anthropometric data, including weight and height in both groups, was measured by a trained healthcare practitioner. Then, the Body Mass Index (BMI) was calculated by dividing weight in kilograms by the square of height in meters. A BMI less than 25 was considered normal, and a BMI of greater than 25 was considered overweight and obese.

Main contributors

All the risk factors were self-reported except for weight, height, and BMI. Tobacco and drug use, drinking habit, BMI group (under and upper 25), history of serious or chronic disease, allergy history, history of cancer in the family, occupation status, doing regular daily exercise, marital status, income sufficiency, and education level as demographic features were extracted from each subject's record. In the nutritional dimension, we asked about key dietary practices from both case and control groups, including the daily number of meals, the habit of adding salt to food, the extent of frying various foodstuffs (such as vegetables, potatoes, onion), methods of preserving vegetables, the types of containers used for bread, vinegar, pickles, and tomato juice, and the kinds of cooking utensils employed.”

Data mining

In this study, the random forest model, which is one of the machine learning models, was used for cancer diagnosis. To train, test and validate the model 75% and 25% of the data were used respectively. Based on the training data, the importance of 17 food predictors (base) for cancer and the non-cancer diagnosis was investigated and the random forest model was fitted; the performance of the model was fitted to the test data using cross-validation and various evaluation criteria such as Matthew's correlation coefficient (MCC), confusion matrix, classification report, and receiver operator characteristic (ROC) curve were reviewed.

Data manipulation was performed in Excel version 2020, while calculations were performed in Python version 3.10, along with pandas, numpy, sklearn, seaborn, and matplotlib libraries. To determine the optimal amount of trees in the Random Forest algorithm, the graph of the discrete error values (obtained from the out-of-bag values for different numbers of trees) was drawn (Fig. 1).

According to the data presented in Fig. 1, it is evident that the minimum error occurs for a range of tree numbers. This information was obtained by inputting the corresponding outputs from the chart into an Excel file. Specifically, for the tree numbers 86, 87, 88, 89, 90, 92, and 93 the minimum error was found to be 0.079. Subsequently, the Random Forest model was assessed using these values, leading to the determination that the optimal number of trees is 93.

Performance was evaluated using 15-fold cross-validation, which resulted in an average accuracy of 92%. In the cross-validation method, the data is randomly divided into k subsets of equal size. The process of training and testing is executed k times so that in the i-th iteration, the subset D_i is considered as the test set, and other subsets are considered as the training set of the model [19]. The learning model is fitted for the training category and its performance is evaluated and validated in the test subset. Therefore, the training and validation process is performed in each iteration.

Model performance was assessed using standard metrics such as accuracy, precision, and AUC (Area Under the ROC Curve) associated with the confusion matrix and the Receiver Operator Characteristic curve. The calculation of the contributions of risk factors in the prediction, specifically the features, was conducted by analyzing feature importance.

Statistical Analysis

Subject characteristics are reported as mean ± Standard Deviation (Mean ± SD) for continuous variables and as frequencies and percentages (%) for categorical variables.

This study was conducted on 252 individuals (126 cases and 126 control) and 58% and 56% were female in case and control groups respectively. The most prevalent cancers in females were breast (19.4%) and stomach (5.6%) and in males were prostate (11.9%) and colon (6.4%). Comparison of age mean at the beginning of the study and BMI between groups showed no significant difference (p-value˃0.05). Additionally, the mean age at the time of cancer diagnosis in the case group has been reported in Table 1.

Also, the participants' main characteristics, including biological and lifestyle features, according to the case and control groups, are presented in Table 2.

Table 1

Participants' age and BMI in case and control groups
	Groups		p-value
	Control	Case
	Mean ± SD	Mean ± SD
Age at the time of the interview	58.42 ± 12.69	58.49 ± 12.72	0.96
Age at the time of diagnosis	-	54.04 ± 12.80
BMI	26.39 ± 4.01	26.04 ± 4.50	0.51

Table 2

Participants' characteristics in terms of case and control groups
		Groups
		Control		Case
		N	%	N	%
Tobacco use	no	65	51.6	94	74.6
Tobacco use	yes	61	48.4	32	25.4
BMI_groups	25 or less	45	35.7	57	45.2
BMI_groups	> 25	81	64.3	69	54.8
History of serious or chronic disease	yes	71	56.3	106	84.1
History of serious or chronic disease	no	55	43.7	20	15.9
Drug use	no	120	95.2	122	96.8
Drug use	yes	6	4.8	4	3.2
Drinking habit	no	105	83.3	126	100.0
Drinking habit	yes	21	16.7	0	0.0
Occupation status	Housewife	38	30.2	66	52.4
	Farmer	20	15.9	19	15.1
	Worker	20	15.9	6	4.8
	Retired	48	38.1	35	27.8
History of cancer in the family	yes	58	46.0	33	26.2
History of cancer in the family	no	68	54.0	93	73.8
marital status	Single	10	7.9	5	4.0
	Married	89	70.6	94	74.6
	Other	27	21.4	27	21.4
Income	I have no income	24	19.0	30	23.8
	Insufficient income	62	49.2	26	20.6
	Enough income but with management	27	21.4	45	35.7
	Enough to cover expenses	7	5.6	20	15.9
	Enough and easily meet the needs without saving	6	4.8	5	4.0
Education level	Illiterate	19	15.1	25	19.8
	Diploma	84	66.7	95	75.4
	University education	23	18.2	6	4.8

Initially, a criterion for evaluating the model's performance, known as the out-of-bag (OOB) evaluation criterion, was computed. This criterion is specifically designed for ensemble learning techniques like Random Forest and serves as an indicator of the model's accuracy when presented with unseen data [20]. To elaborate, each decision tree within the Random Forest ensemble is trained on a bootstrap sample derived from the original data, resulting in the exclusion of certain data points during the training process, referred to as out-of-bag (OOB) instances.

After training each tree, the OOB samples were used to estimate the performance of the model by calculating the OOB score, which was found to be 0.90 based on the studied data, which indicates that the Random Forest model achieved approximately 90% accuracy in the OOB samples.

To gain a comprehensive understanding of the model's performance with 93 trees accuracy, precision, recall, and F1 score were calculated (Table 3). The value of Matthew's correlation coefficient (MCC) is a high value, which measures the difference between the predicted and actual values. Considering the high value of MCC, it can be concluded that many cases are predicted correctly and therefore, it is normal for other evaluation indicators to be high.

Table 3

Performance Evaluation Metrics for Random Forest Model
Indices	AUC	Accuracy	F1	Precision	Recall	MCC
Values	0.97	0.97	0.97	0.94	1	0.939
AUC = Area Under the ROC Curve, MCC = Matthew's correlation coefficient

According to the computational findings, it can be stated with certainty that the Random Forest model is very powerful in accurately predicting the occurrence of cancer. In other words, the Random Forest model exhibits a high level of discrimination, as evidenced by its AUC value of approximately 0.97 which the ROC is shown in Fig. 2.

To assess the efficacy of the developed classification model, the confusion matrix of the test data was acquired, as presented in Table 4. In other words, by utilizing said confusion matrix, the quantities of true positive (TP), true negative (TN), false positive (FP), and false negative (FN) test outcomes were established.

Table 4

Confusion matrix
		predicted
Actual		cancerous	Healthy
	cancerous	33	0
	Healthy	2	29

By recording the information related to the characteristics effective in the occurrence of cancer in an Excel file, it was found that 68 features, including storing Vegetables in the freezer, roasting Potatoes, Eggplants, and Squash, etc., have the greatest effect. To increase the values of the features and to have the same interval, they were scaled using the minimax method.

In addition, for quantitative understanding, Fig. 3 shows the 10 features that have high scores. Features with larger values on the x-axis are considered more important in predicting the target variable which was cancer as the main outcome in this study. Accordingly, freezing vegetables, frying potatoes, eggplants, and squash and container type for storing bread are respectively the most important predictors of causing cancer.

Table 5 Top 10 importance features with normalized importance features

The most important features for the total of 252 subjects (126 cases and 126 controls) have been depicted in Fig. 4. Points are colored according to the value of the feature, red for high value and blue for low value. Points with a more skewed distribution seem to be more important features.

In the analysis of this model, it is observed that people with cancer (cases) store vegetables in the freezer or fried form in the fridge, use fried potatoes, eggplants, and squash in gold form, store bread in plastic bags and use plastic dishes to store tomato juice and pickles. These features have predicted cancer in the case group (red dots in the diagram). On the contrary, these characteristics have shown reverse distribution in the control group, so can be considered behaviors with less practice in the late group.

Cancer development is a multifactorial disorder rooted in environmental exposures such as diet, pollution, tobacco use, and physical activity, besides genetic predisposition. The current lack of complete treatments for the eradication of cancers has boosted the search for early markers or predictors of developing neoplasms in people.

Our study aimed to develop a Random Forest Classifier for predicting cancer in Iranian males and females using dietary habits. We used more than fifty lifestyle-related risk factors obtained from 252 participants. Although comprehensive data was collected, we decided to neglect some features in the model whose effects on cancer development have been proven for many years [20, 21], to focus and emphasize on probable hidden aspects that have not been modified yet despite their approved damaging role.

We found 4 risk factors that are strong predictors of cancer, the use of frozen vegetables, fried vegetables in the fridge, gold-fried potatoes, eggplants and squash, plastic bags to store bread, and plastic containers for conserving tomato juice and pickles.

According to the model explanation, vegetable storage in the freezer was distinguished as a key factor in cancer prediction. In comparison, no evidence suggests that frozen foods cause cancer. Indeed, refrigeration may even help reduce cancer risk by preventing the growth of carcinogen microorganisms [22]. Anyway, according to our proposed model, substituting frozen vegetables with dried ones could be a valuable piece of advice in protecting against cancers and going on a healthier diet. It can be explained by being willing to consume fewer fresh vegetables and salads when there are frozen kinds of stuff that are less in antioxidants content [23].

Frying, a popular food processing method, was shown to be carcinogenic in the present study. Abundant studies have confirmed that the production of cytotoxic, neurotoxic, and mutagenic complexes, such as malondialdehyde (MDA), a peroxidation product of frying oils, is a contributing factor [24]. Moreover, vegetable oils when frequently treated with high temperatures produce various byproducts, such as polycyclic aromatic hydrocarbons (PAHs), well-known to be poisonous and toxic [25]. Our findings align with existing data, indicating that an increased consumption of deep-fried foods increases the chance of cancer development.

Preserving bread in plastic bags between 4 top-ranked features grabs excellent attention. As a cultural habit in Iran, people put fresh and warm (even hot) bread in plastic bags and package it in these covers until they use it. Our model shows that this behavior is one of the most important features in cancer incidence while it is modifiable by encouraging people to use fabric or paper-based pockets to keep bread, especially when it is hot. Moreover, reusing plastic bags several times, which is common here, may signify the harmful effect of this kind of package. It is suggested that the chemicals in the food containers can transfer to the food and unintentionally become part of the human diet [26]. More than 148 chemicals have been recognized that possess toxicological and endocrine-disrupting characteristics, accumulate in the body, and are associated with cancer, including melamine, BisPhenol A, and DBP which are approved as a carcinogen in prostate and breast cancer [27, 28].

Moreover, in some cases, unique compounds with antimicrobial properties are added to the components of plastic containers to keep food or water fresher. Silver nanoparticles, as one of these compounds, migrate to food in an acidic environment, similar to the previously mentioned compounds [29]. However, not many studies have been conducted on the intensity and size of nano-silver particles transferred from container to food. Still, it can be considered as a potential issue in preserving acidic liquids such as vinegar or tomato juice in plastic containers.

Random Forest, a highly favored machine learning algorithm, shows strong performance, especially in datasets of smaller to medium sizes [30]. Significantly, in the context of this study, the algorithm of choice for automated feature selection is the random decision forest that effectively addresses the essential issue of feature selection, providing insights into the most pivotal features for optimizing prediction accuracy. Our results showed that 10-based features emerged as the best model as it gave 97% accuracy, 94% precision, and 97% AUC in predicting cancer which are high values.

However, the Random Forest model has been applied in various surveys to predict situations such as cancer, obesity, heart diseases, and stroke [31–34], to the best of our knowledge, no study investigated the factors we considered in cancer development through this model. The presented algorithm seems to be working as a prognosis support tool that aids clinicians in emphasizing factors that have not been given much attention before but are serious risk elements, according to our results. It is essential to mention that the present algorithm uses inexpensive and non-invasive procedures such as sociodemographic, lifestyle, and self-reported features, which makes its application easy.

The main limitation of our study is the sample size. However, we included all the eligible people with diagnosed cancer to enter our study, they did not exceed 126 subjects in the case group, and this points out that the results may not be generalizable. In addition, incomplete data on some questions forced us to omit them. Future studies with greater sample sizes and more complete answers to questionnaires will make the results more generalizable.

We studied the total population with cancer in three cities in the Semnan province of Iran. Our target was to establish a predictive model through machine learning algorithms that can allow cancer to be predicted using modifiable risk factors that are common between various types of cancers. A Random Forest Classifier was developed successfully with high predictive value at predicting cancer and valuable information which emphasizes paying attention to those risk factors that are overlooked even in health care system recommendations. Despite the extensive information provided to individuals with cancer to improve their quality of life, seemingly simple instructions such as altering food containers and guidelines on food preparation and storage are often overlooked and ignored which highlights a potential gap in attention to the comprehensive care of cancer patients. Addressing and emphasizing the significance of these basic recommendations may enhance overall patient well-being and treatment outcomes.

Acknowledgments

Our sincere gratitude is extended to Dr. Hamidreza Marateb of the Biomedical Engineering Research Centre (CREB) at the Universitat Politècnica de Catalunya-Barcelona Tech, as well as to Dr. Marjan Mansourian of the Department of Epidemiology and Biostatistics at the School of Health, Isfahan University of Medical Sciences, for their insightful comments on this manuscript. Additionally, we would like to express our appreciation to all the participants who contributed to this study and the healthcare providers who assisted in the completion of the dataset.

Author contributions

MH prepared the Methodology and conducted the Formal analysis.

MM conceptualized the project, administered it, and supervised it.

MMB and AS carried out the project, gathered data, and conducted participant interviews.

MD cleaned the data and wrote and edited the first draft.

MP mined the data and reviewed and edited the first draft.

MN conceptualized the project, wrote the original draft, and provided supervision.

All authors reviewed the manuscript.

Funding

The work performed for this article received funding from Deputy of Research and Technology of Semnan University of Medical Sciences under the Grant Agreement No. IR.SEMUMS.REC.1399.312.

Competing interests The authors have no relevant financial or non-financial interests to disclose.

Ethical approval This study was approved by the Ethics Committee (IR.SEMUMS.REC.1399.312) and informed consent was obtained from all participants after explaining the objectives and details of the study.

Data availability The data that support the findings of this study are available from the corresponding author, MN, upon reasonable request.

Sung H, Ferlay J, Siegel RL, et al (2021) Global Cancer Statistics 2020: GLOBOCAN Estimates of Incidence and Mortality Worldwide for 36 Cancers in 185 Countries. CA Cancer J Clin 71:209–249. doi: 10.3322/caac.21660
Tran KB, Lang JJ, Compton K, et al (2022) The global burden of cancer attributable to risk factors, 2010–19: a systematic analysis for the Global Burden of Disease Study 2019. Lancet 400:563–591. doi: 10.1016/S0140-6736(22)01438-6
Ferlay J, Ervik M, Lam F, et al (2018) Global cancer observatory: cancer today. Lyon, Fr Int agency Res cancer 3:2019.
Republic I (2019) Iran (Islamic Republic of). World Popul Policies 2015 967:292–293. doi: 10.18356/5d3c4e28-en
Roshandel G, Ferlay J, Ghanbari-Motlagh A, et al (2021) Cancer in Iran 2008 to 2025: Recent incidence trends and short-term predictions of the future burden. Int J Cancer 149:594–605. doi: 10.1002/ijc.33574
American Cancer S (2023) American Cancer Society. Cancer Prevention & Early Detection Facts & Figures 2023-2024.
Danaei G, Vander Hoorn S, Lopez AD, et al (2005) Causes of cancer in the world: comparative risk assessment of nine behavioural and environmental risk factors. Lancet (London, England) 366:1784–1793. doi: 10.1016/S0140-6736(05)67725-2
Islami F, Goding Sauer A, Miller KD, et al (2018) Proportion and number of cancer cases and deaths attributable to potentially modifiable risk factors in the United States. CA Cancer J Clin 68:31–54. doi: 10.3322/caac.21440
Brown KF, Rumgay H, Dunlop C, et al (2018) The fraction of cancer attributable to modifiable risk factors in England, Wales, Scotland, Northern Ireland, and the United Kingdom in 2015. Br J Cancer 118:1130–1141. doi: 10.1038/s41416-018-0029-6
Grosso G, Bella F, Godos J, et al (2017) Possible role of diet in cancer: Systematic review and multiple meta-analyses of dietary patterns, lifestyle factors, and cancer risk. Nutr Rev 75:405–419. doi: 10.1093/nutrit/nux012
Obermeyer Z, Emanuel EJ (2016) Predicting the Future - Big Data, Machine Learning, and Clinical Medicine. N Engl J Med 375:1216–1219. doi: 10.1056/NEJMp1606181
Bellazzi R, Zupan B (2008) Predictive data mining in clinical medicine: Current issues and guidelines. Int J Med Inform 77:81–97. doi: https://doi.org/10.1016/j.ijmedinf.2006.11.006
Koh HC, Tan G (2011) Data mining applications in healthcare. J Healthc Inf Manag 19:65.
Octaviani TL, Rustam Z (2019) Random forest for breast cancer prediction. AIP Conf Proc 2168:20050. doi: 10.1063/1.5132477
Wang S, Wang Y, Wang D, et al (2020) An improved random forest-based rule extraction method for breast cancer diagnosis. Appl Soft Comput 86:105941.
S SN (2023) Prediction of Breast Cancer Through Random Forest. Curr Med imaging 19:1144–1155. doi: 10.2174/1573405618666220930150625
van Zutphen M, van Duijnhoven FJB, Wesselink E, et al (2021) Identification of Lifestyle Behaviors Associated with Recurrence and Survival in Colorectal Cancer Patients Using Random Survival Forests. Cancers (Basel). doi: 10.3390/cancers13102442
Chen W, Xia C, Zheng R, et al Articles Disparities by province , age , and sex in site-specific cancer burden attributable to 23 potentially modifiable risk factors in China : a comparative risk assessment. Lancet Glob Heal 7:e257–e269. doi: 10.1016/S2214-109X(18)30488-1
Ha J, Kambe M, Pe J (2011) Data Mining: Concepts and Techniques. Data Min Concepts Tech. doi: 10.1016/C2009-0-61819-5
Katzke VA, Kaaks R, Kühn T (2015) Lifestyle and Cancer Risk. Cancer J. 21:
McCormack VA, Boffetta P (2011) Today’s lifestyles, tomorrow’s cancers: trends in lifestyle risk factors for cancer in low- and middle-income countries. Ann Oncol 22:2349–2357. doi: https://doi.org/10.1093/annonc/mdq763
Yan S, Gan Y, Song X, et al (2018) Association between refrigerator use and the risk of gastric cancer: A systematic review and meta-analysis of observational studies. PLoS One 13:e0203120. doi: 10.1371/journal.pone.0203120
Coggon D, Barker DJ, Cole RB, Nelson M (1989) Stomach cancer and food storage. J Natl Cancer Inst 81:1178–1182. doi: 10.1093/jnci/81.15.1178
Ma L, He Q, Qiu Y, et al (2021) Food matrixes play a key role in the distribution of contaminants of lipid origin: A case study of malondialdehyde formation in vegetable oils during deep-frying. Food Chem 347:129080. doi: https://doi.org/10.1016/j.foodchem.2021.129080
Ganesan K, Xu B (2020) Deep frying cooking oils promote the high risk of metastases in the breast-A critical review. Food Chem Toxicol 144:111648. doi: https://doi.org/10.1016/j.fct.2020.111648
Muncke J (2021) Tackling the toxics in plastics packaging. PLOS Biol 19:e3000961.
Weber Lozada K, Keri RA (2011) Bisphenol A Increases Mammary Cancer Risk in Two Distinct Mouse Models of Breast Cancer1. Biol Reprod 85:490–497. doi: 10.1095/biolreprod.110.090431
Tse LA, Lee PMY, Ho WM, et al (2017) Bisphenol A and other environmental risk factors for prostate cancer in Hong Kong. Environ Int 107:1–7. doi: https://doi.org/10.1016/j.envint.2017.06.012
Mackevica A, Olsson ME, Hansen SF (2016) Silver nanoparticle release from commercially available plastic food containers into food simulants. J Nanoparticle Res 18:5. doi: 10.1007/s11051-015-3313-x
Gómez-Ramírez J, Ávila-Villanueva M, Fernández-Blázquez MÁ (2020) Selecting the most important self-assessed features for predicting conversion to mild cognitive impairment with random forest and permutation-based methods. Sci Rep 10:1–15. doi: 10.1038/s41598-020-77296-4
Biesbroek S, Van Der A. DL, Brosens MCC, et al (2015) Identifying cardiovascular risk factor-related dietary patterns with reduced rank regression and random forest in the EPIC-NL cohort. Am J Clin Nutr 102:146–154. doi: 10.3945/ajcn.114.092288
Shao Z, Xiang Y, Zhu Y, et al (2020) Influences of Daily Life Habits on Risk Factors of Stroke Based on Decision Tree and Correlation Matrix. Comput Math Methods Med. doi: 10.1155/2020/3217356
Husain W, Wei LJ, Cheng SL, Zakaria N (2011) Application of data mining techniques in a personalized diet recommendation system for cancer patients. 2011 IEEE Colloq Humanit Sci Eng CHUSER 2011 239–244. doi: 10.1109/CHUSER.2011.6163724
Tatoli R, Lampignano L, Donghia R, et al (2022) Dietary Customs and Social Deprivation in an Aging Population From Southern Italy: A Machine Learning Approach. Front Nutr 9:1–8. doi: 10.3389/fnut.2022.811076

No competing interests reported.

Application of Random Forest Model in Cancer Risk Prognosis: A Role for Dietary Habits

Status:

Version 1

Abstract

Figures

Background

Methods

Subjects

Data gathering

Main contributors

Data mining

Statistical Analysis

Results

Discussion

Conclusions

Declarations

References

Additional Declarations

Status:

Version 1