Search

This research focused on analyzing and classifying the survival of some breast cancer in-patients in Kwara State, Nigeria. The relationships between some variables that were measured to determine the survival of breast cancer in a cross-sectional study in Nigeria were explored. Data detailing information on gender, age, mode of diagnosis, length of stay, location of breast cancer and the outcome of the treatment on 300 patients admitted based on breast cancer issues were collected at the University of Ilorin Teaching Hospital (UITH), Ilorin, Nigeria between year 2011 and 2016. Descriptive analyses were performed; cluster analysis was performed using k-mean clustering method with the aid of Orange software. Analysis of the cluster is done to reveal the latent patterns that could lead to a better understanding on how some variables could combine to determine the extent of breast cancer survival in Nigeria The scatter plot was used for data visualization while the Silhouette plot showed validates the choice of using only three clusters. Breast cancer survival is associated with the age bracket of the patients. Also, breast cancer survival is dependent on the location of the cancer. In addition, mode of diagnosis was found to be associated with the location of the breast cancer. Furthermore, k-clustering algorithm identified three distinct clusters. The low Silhouette values of obtained in both plots showed that only few clusters are available for this type of data. Louvain clustering revealed the latent patterns that depicts the prevalence of breast cancer in Nigeria. [ABSTRACT FROM AUTHOR]

Breast cancer is one of the leading causes of death in females and survival depends on early diagnosis and treatment. This paper applied machine learning techniques in prediction of breast cancer survival (dead or alive) using age, sex, length of stay, mode of diagnosis and location of cancer as predictors (independent variables). The data was obtained from the outpatient department of the University of Ilorin Teaching Hospital, Ilorin, Nigeria. The sample size of 300 consists of 175 females and 25 males who were admitted at the hospital and treated for breast cancer. The patients were later discharged or died. Adaptive boosting (AdaBoost) performed best out of the data mining models used in the classification in all the three cases where the target class is average over classes, alive or dead. The AdaBoost performed best with the classification accuracy and area under curve (AUC) of 98.3% and 99.9% respectively. Furthermore, a probe on the prediction by AdaBoost showed that the probability of dead due to breast cancer is 0.47, which the length of stay hugely contributed to the high probability, location of breast cancer and mode of diagnosis contributed minimally while age and sex contributed insignificantly. The high probability of breast cancer mortality predicted in this paper is a call for concern as early detection of breast cancer, routine breast examination and breast cancer awareness are crucial in increasing the probability of survival. The results can be used to design a decision support system that can increase the chances of breast cancer survival. [ABSTRACT FROM AUTHOR]

Malaria is a leading cause of death in Nigeria and remains a public health concern because of the increasing resistance of the disease to antimalarial drugs. Pregnant women and children under five years of age are the most vulnerable. Efforts to eradicate malaria is often frustrated due to some various sociodemographic factors and medical factors. One of the vital therapeutic factors is misdiagnosis. Hence, the paper applied different data mining models to diagnose malaria using fifteen symptoms of patients that attended a hospital in Nigeria. The data were obtained from a peer reviewed data article that comprises 337 subjects at Federal Polytechnic Ilaro Medical Centre, Ogun State. The independent variables are 15 symptoms, age, and sex, while the target or dependent is the outcome. The outcome is the result of the diagnosis, which is positive for negative for malaria. Eight machine learning tools were applied to the data on the Orange Software platform. Weak non-significant correlations were obtained between the 15 symptoms and the outcome, and hence no pattern was observed. However, the application of data mining tools revealed a hidden pattern that correctly predicted the outcome using the subjects' symptoms, age, and sex. 6 out of the 8 machine learning models were adjudged to perform well using different performance metrics. The Adaptive boosting model gave a percent 100% precision in the classification, and logistic regression was the least. Furthermore, a percent performance of Adaboost implies that the model correctly predicted all the 221 true negatives and 116 true positives with a misclassification (misdiagnosis) of zero. Classify using only the 15 symptoms reduced the predictive accuracy of the 6 models. Nevertheless, Adaboost performance was the best with a classification accuracy of 98.2%, precision of 96.6%, and an error rate of just 1.8%. Again, logistic regression performance was the least. The present work has presented a strong relationship between age and sex and the outcome. Adaboost model can be used to design decision support systems or rapid diagnostic tools that utilise the internet or mobile devices as platforms in the diagnosis of malaria. The application of the present work as potentials in reduction of misdiagnosis incidences, reducing the mortality due to malaria and improving the overall public health of people residing in malaria endemic areas. [ABSTRACT FROM AUTHOR]