Autor: |
Avni Mehta, Richa Nayak, Yasha Hasija |
Rok vydání: |
2020 |
Předmět: |
|
Zdroj: |
2020 IEEE International Conference on Computing, Power and Communication Technologies (GUCON). |
DOI: |
10.1109/gucon48875.2020.9231223 |
Popis: |
Melanoma is a highly prevalent dermatological disease and a life-threatening form of skin cancer. Lifestyle factors have been observed to influence melanoma risk. The relationship between increased BMI and melanoma incidence has been investigated using epidemiological, in our previous studies. In order to study relationships between melanoma and other modifiable lifestyle factors such as alcohol consumption, smoking, sunscreen application, and use of tanning devices, a meta-analysis aided by machine learning technique was carried out. PubMed database was searched carefully to sort out literature pertaining to the associations of these lifestyle factors with melanoma risk. Meta-analyses were carried out using a software called Review Manager 5.3. It gave risk ratios and 95% confidence intervals as results. Analysis of these data revealed weak positive relationship between alcohol consumption $(\text{OR} =1.46;\ 95\%\ \text{CI} =1.32-1.62)$ and use of tanning devices $(\text{OR}=1.36;\ 95\%\ \text{CI}=1.20-1.53)$ . Negative associations were found between smoking $(\text{OR}=0.78;\ 95\%\ \text{CI} =0.67-0.92)$ and sunscreen application $(\text{OR}=0.5;\ 95\%\ \text{CI}= 0.44-0.57)$ . These results were matched, and association trends were confirmed with the rule induction results obtained by applying Naive Bayes model to the data of each lifestyle factor. All the data from the five factors were pooled in together to create a master datasheet, and machine learning was performed on it to generate a predictive model for melanoma risk. The results were validated through a test split (ratio 0.7:0.3) and cross-validation as well. The accuracy was observed to be 70.23% and 70.35% + 0.79%. The functioning of the model was tested on an unlabeled dataset. |
Databáze: |
OpenAIRE |
Externí odkaz: |
|