OCHOR, EMMANUEL CHERECHI (2026) DEVELOPMENT OF A BREAST CANCER RISK PREDICTION SYSTEM USING ARTIFICIAL NEURAL NETWORKS (ANNs). Other thesis, Godfrey Okoye University, Enugu Nigeria.
![[img]](https://eprints.gouni.edu.ng/style/images/fileicons/text.png) |
Text
Ochor Emmanuel - Full Work.pdf Download (2MB) |
Abstract
Based on data showing that more than 2.3 million new cases were recorded in 2022, Breast cancer has topped the list of the most commonly diagnosed cancers among women internationally. This data implies practically, that smart tools for accessing the risk of breast cancer have become very important, more especially in locations where there is limited or no groundwork for conventionally screening these women. Convention risk models rely on narrow predictor sets and population-specific assumptions, even though they are clinically established. An example is the Gail/BCRAT tool, which cannot capture complex non-linear interactions among clinical variables. Also, many studies in this field focus on comparing algorithms and fail to develop a practical software solution. This research creates a breast cancer classification model based on biomarkers utilizing an artificial neural network (ANN). The input/output platform is a web-based prototype. A shallow multilayer perceptron model with two hidden layers was developed, and it was trained using the Adam optimizer, binary cross-entropy loss, dropout regularization, and the early stopping technique on a publicly available Breast Cancer Coimbra dataset from the UCI Machine Learning Repository. The dataset consists of 116 casecontrol records and each record is described by nine anthropometric and serum biomarkers. After training, it was evaluated against logistic regression, support vector machine, and random forest baselines. On the held-out test split, the ANN has attained an accuracy of 82.61%, and precision and recall of 84.62%, specificity of 80.00%, and an ROC-AUC of 0.86; the ANN was better than the other three baselines. Still, these figures may have significant variation due to the limited data sample and they should be interpreted as only indicative. The trained model with the processing scaler used was a part of a Flask web application that Beyond user authentication and structured data entry features, also supports the interpretation of results based on probabilities, keeps users' assessment history, and generates reports. The findings show that a shallow ANN can extract significant relationships in small structured biomarker data and it can also be part of a browser-based tool that can be used by people very easily. This tool could be a step in the direction of modelling research and decision support that can be deployed. Still, it is merely a proof-of-concept system which identifies existing cases and controls rather than predicting future risk and should not be equated with a clinical diagnostic instrument; the next steps in research should be focused on larger and prospective datasets, external validation, cross- validated performance estimates, and probability calibration before considering any real- world use.
| Item Type: | Thesis (Other) |
|---|---|
| Subjects: | Q Science > Q Science (General) |
| Divisions: | Faculty of Computing And Information Technology (FACIT) |
| Depositing User: | Cynthia Ugwuoti |
| Date Deposited: | 31 Jul 2026 09:19 |
| Last Modified: | 31 Jul 2026 09:19 |
| URI: | http://eprints.gouni.edu.ng/id/eprint/6110 |
Actions (login required)
 |
View Item |
