Predicting determinants of modern contraceptive use among reproductive-age women in Ethiopia using machine learning algorithm: Evidence from the Performance Monitoring and Accountability (PMA) Survey 2019 dataset

Introduction

Approximately 40% of women worldwide report having unwanted pregnancies.¹ In developing nations, an estimated 214 million women of reproductive age who want to avoid pregnancy do not use any method of contraception, and in 2017, the use of modern contraceptives is expected to have saved 308 million unwanted pregnancies.² Although this expectation has not yet been met in Sub-Saharan Africa, it has been observed in many other regions of the world, particularly in Asia and Latin America.³ Sub-Saharan Africa as a whole still has the highest fertility rate worldwide.³ According to reports from 2012 and 2017, only a small percentage of women in Africa used modern contraceptives, with estimates of 23.9% and 28.5%, respectively.⁴ In a recent large population-based study, the prevalence of modern contraceptive use was estimated to be 26% among women of reproductive age in 20 African countries, with country-specific variations ranging from 6% in Guinea to 62% in Zimbabwe.⁵

Modern contraceptives (MCs) are widely accepted to limit rapid population increases, especially in developing nations, and have been proven to be an effective approach for reducing fertility.^6,7 The use of MCs provides a number of advantages. These include birth spacing, lowering the number of unexpected or unwanted pregnancies, preventing unsafe abortions, enhancing maternal health, lowering infant mortality, and preventing STDs.⁸ The decrease in poverty, the expansion of women’s educational possibilities, and the ensuing sustainable population growth and economic development for nations are among the non-health benefits that have been identified.⁹

According to a study conducted in Vietnam, the association between a woman’s age and current contraceptive methods is shaped like an inverted “U.” Although the chance of using contraception was low among women aged 15 to 24, it was greater among those in the 25 to 35 age group and lowest among those aged 35 and older.¹⁰ The adoption and use of modern contraceptive methods by women can also be influenced by their degree of education. Employed women with higher educational levels had a noticeably greater likelihood of using contraceptives than illiterate women did, according to a study of the prevalence and determinants of contraceptive usage among employed and jobless women.¹¹ According to a Nigerian study, women with higher (tertiary) education are four times more likely to use modern contraceptives than women with lower educational attainment.¹² In a similar vein, wives of highly educated men were more likely to accept and support the use of modern contraceptive techniques.¹³

Numerous investigations have shown a strong correlation between place of residence and the use of modern contraceptive methods.^14,15 Women in urban regions are more likely than women in rural areas to use modern contraceptive techniques, although the majority of people live in rural areas.¹⁶ In relation to modern contraceptives, a woman’s wealth index and type of wage impact her financial status as well as her accessibility and affordability.¹⁷ The acceptance and use of a modern contraceptive by a woman can be influenced by her marital status.¹⁸ The choice and use of modern contraceptive methods have been linked to cultural influences, religion, and information sources, all of which have an impact on women’s decisions.¹⁹

To pinpoint the causes of the low use of modern contraceptives, numerous studies have been conducted in Ethiopia and other countries.^14,15 Their findings suggest that low use of contraceptives is responsible for the high fertility rates in sub-Saharan African nations, which have an adverse impact on early childbirth, high infant and maternal mortality, and a host of other socioeconomic factors.^20,21 Using traditional regression models, earlier research conducted in this country demonstrated the effects of socioeconomic and demographic factors related to the use of modern contraceptives, which became less accurate as the number of variables used and the potential correlations increased.^22,23 These traditional models usually involve problems involving multidisciplinary relationships between variables and many factors.^24,25 Machine learning (ML) thus offers an efficient method for discovering pertinent characteristics connected to certain health outcomes for public health research compared to traditional models.^24,26,27 Therefore, this study aimed to assess the determinants of modern contraceptive use among reproductive-aged women in Ethiopia using six widely used machine learning (ML) algorithms. This study sought to determine and identify consistent determinants and others of modern contraceptive use using the Performance Monitoring and Accountability (PMA) Survey 2019 dataset for currently non-pregnant reproductive-age women in Ethiopia. The most influential and consistent determinants identified based on these findings will serve as priority intervention areas for which the Ethiopian Ministry of Health and other health partners can concentrate to improve the use of modern contraceptives in Ethiopia.

Method

Data processing and analysis

The R programming language (version 3.6.0) and the caret package (Kuhn, 2020) were used to perform the data processing and analysis. We used six widely accepted machine learning algorithms to predict the determinants of modern contraceptive use in Ethiopia and compared the results of the best algorithm to the results of the traditional logistic regression model. The KNN model is chosen based on its ability to detect linear and nonlinear boundaries between groups. The K value represents the number of nearest neighbors and is the core deciding factor in this classifier. The RF model is used in machine learning situations because it is highly flexible and provides good predictive performance. It produces ensemble predictions that are more accurate than any of the individual predictions. The naive Bayes algorithm is a supervised machine learning algorithm that uses the Bayes theorem for classification and prediction. It has an incremental learning behavior and is not affected by training time. Logistic regression is a statistical model that is used to classify and predict different health parameters, where the target variable is dichotomous and the independent variables are independent of each other. The general framework used in the literature^31,32 based on Yufeng Guo’s 7 machine learning steps was used in this study. The framework describes the seven steps in supervised machine learning, which are as follows: data collection, data preparation, model selection, model training, model evaluation, parameter tuning, and prediction.

Data source/collection

The dataset for this study is available on the PMA Survey website and can be obtained upon formal request. A weighted sample of 8837 reproductive-age women was included in the data. The datasets analyzed in the current study are available in the PMA repository, https://www.pmadata.org/data/available-datasets .

Data preparation: The data preprocessing techniques used in this investigation included data cleaning, data splitting, feature engineering, and dimensionality reduction. After the data were extracted, data cleaning was performed, which included finding and removing outliers from the dataset, as well as resolving missing values and uneven categories in the resulting variable. This study employed the k-nearest neighbors (KNN) imputation approach to compensate for the missing values in the dataset related to the independent variables. KNN imputation with the mouse package was used since it is a more reliable and sensitive technique for the imputation of missing values.^27,33

The raw data were transformed into features that more accurately depicted the underlying issue for the predictive models, improving the accuracy of the unobserved data. Hence, among other feature engineering techniques, label encoding for coding each category of variables as a number and encoding categorical variables into numeric values for nominal variables were carried out. The process of dimension reduction was used to decrease the number of input variables for the predictive model. Having fewer input variables might lead to a simpler predictive model, which can perform better when generating predictions on new information. Using statistics to assess the link between the independent input components and the output variable and selecting the most important independent variables, feature selection and feature extraction were utilized to forecast the target variable. This technique has frequently been used in earlier public health research to identify the factors and/or predictors of different health outcomes.^24,34

This study utilized a standard 80/20 split approach, where 80% of the data were used for training and the remaining 20% were used for model testing. The model was trained using the tenfold cross-validation method, which does not waste much data; this approach is highly beneficial when the number of samples is limited.²⁴ To train the prediction function, K-fold separates all the observations into equal-sized groups of samples termed folds and k-1 folds. The fold that is left out is then used for testing k times repeatedly.²⁷ The average of the results calculated in the loop serves as the k-fold cross-validation performance measure.

Model selection: Appropriate models were selected for training after the data were prepared and split into training and testing sets. The task was a binary classifier because the outcome variable was categorical, and suitable classifiers needed to be chosen to make the prediction. Since modern contraceptive use was divided into two mutually exclusive categories (use or not use), the dataset employed in the analysis falls under the category of binary categorization. Hence, in this study, we used six widely used machine learning (ML) algorithms,²⁷ logistic regression (LR), random forest (RF), K-nearest neighbors (KNN), extreme gradient boosting (XGBoost), naive Bayes (NB), and support vector machines (SVMs), to predict determinants of modern contraceptive use among reproductive-age women in Ethiopia and compared the results of the traditional logistic regression model to the results of the best algorithm for identifying the new features influencing the outcome of interest.

Model training: The selected classifiers were trained using prepared data after model selection, and their performances were compared via tenfold cross-validation. Following this comparison, the top predictive model was chosen, and it was trained with balanced training data to make the final prediction on hypothetical test data.²⁷

Model evaluation: Testing the model’s performance on never-before-seen data that were set aside for this purpose during data splitting can help determine how well the model works after it has been trained. One popular technique for evaluating the effectiveness of a classification model is the confusion matrix, which is a straightforward cross-tabulation of the actual and predicted categories for the outcome variable. The performance criteria, which include the overall accuracy, precision, recall, and F1 score and were employed in this study to evaluate the effectiveness of the selected classifiers, can be calculated using confusion metrics. Additionally, the performance of the ML models was assessed using receiver operating characteristic (ROC) curves, and the value 0.5 = no discrimination. 0.5-0.7 = Poor discrimination. 0.7-0.8 = Acceptable discrimination. 0.8-0.9 = Excellent discrimination.³⁴

Hyperparameter tuning: To better understand the possibility of obtaining the optimal values and avoid unnecessary computations for combinations of nonperforming parameters when searching for the optimal parameter settings. Grid search, random search, and Bayesian optimization were used to formulate hyperparameter optimization to improve the speed and quality of the learning process, and we attempted to incrementally adjust the parameters of our model to improve its performance.³⁵

Making prediction: All the aforementioned activities occur in this stage, which is the last stage in machine learning methodology. By using independent variables as a framework, prediction involves estimating the outcome variable. In this instance, modern contraceptive use was established using crucial factors that were discovered along the route. The best-performing classifier with a specified accuracy was used to predict whether a woman would use modern contraceptive services given various factors. The overall workflow of the methodology is shown below ( Figure 1).

Figure 1. Overview of methodologies.

Results

Modern contraceptive use

Among the study participants, only 24% {95% CI (23.1%, 24.9%)} used modern contraceptive methods. Most (1,204, 13.6%) of the modern contraceptive users had a secondary education ( Figure 2).

Figure 2. Modern contraceptive use among reproductive-age women in Ethiopia: Evidence from the Performance Monitoring and Accountability (PMA) Survey 2019 dataset.

Machine learning analysis of modern contraceptive use among reproductive-age individuals in Ethiopia

Feature selection

Feature selection is crucial for determining the most important predictors of an outcome variable, similar to how p values and t statistics are used in most traditional statistical methods, to determine which variables are significant. Accordingly, when using an ensemble model to predict an outcome, feature importance measures how significant a feature is on average concerning other features.

The most significant predictors of modern contraceptive use according to the extreme gradient boost feature importance results were age at first family planning use (20 years and above), marital status (single), partner/husband feelings about family planning (Approval from partner) and relationship with the head of household (having a wife relationship to the head). In addition, age (36 to 49 years), health care providers advise about the use of FP methods during health facility visit, perception about the FP side effects might disrupt their relationship and household size (5 and above) were other key predictors. The length of the bars on the x-axis, which represents the relative significance of the independent variables in predicting the use of modern contraceptives, shows this. The longer the bar is, the more significant the trait is in determining whether a woman utilizes modern contraception ( Figure 3).

Figure 3. Feature importance plot using the XGBoost model for predictors of modern contraceptive use in Ethiopia.

Evidence from the Performance Monitoring and Accountability (PMA) Survey 2019 dataset.

Note: ageat1stfpuse2 = Age at first family planning use of greater than 30 year; ageatlstfpusel = Age at first family planning use of between 20 and 30 years, marital_status5 = being Single, partner_fp_feeling3 = having FP use approval from partner, relationship2 = having a wife relationship to head of the household, newage3 = being in age of between 36 to 49 years, newage2 = being in age of between 26 to 35 years, facility_fp _discussionl = having discussion on FP method with health professionals during health facility visit, wge_fp_aut_sideeffects_disrupt5 = FP side effects will not disrupt my relationship, num_hh_ members6 = having family size of greater than 5 house hold members.

Model selection

After the data were prepared and divided into training and testing datasets, as indicated in Table 4, the appropriate models were chosen for the training dataset. To predict the factors that influence the use of modern contraceptives among reproductive-age women in Ethiopia, we used six commonly used machine learning (ML) algorithms, namely, logistic regression (LR), random forest (RF), K-nearest neighbors (KNN), extreme gradient boosting (XGBoost), naive Bayes (NB), and support vector machines (SVMs). We chose the best model from among these machine learning approaches based on its higher level of accuracy ( Table 4).

Table 4. Model accuracy metrics for all the models evaluated on the training data.

Algorithms	Minimum	1st Quartile	Median	Mean	3rd Quartile	Maximum
Logistic Regression	0.788	0.798	0.802	0.801	0.804	0.809
Random Forest	0.795	0.796	0.801	0.800	0.805	0.807
XGBoost	0.790	0.803	0.809	0.806	0.812	0.814
KNN	0.769	0.778	0.782	0.785	0.790	0.806
Naive Bayes	0.758	0.793	0.799	0.797	0.807	0.814
Linear SVM	0.792	0.796	0.797	0.800	0.806	0.811

A figure based on the mean level of accuracy serves as the representation for the summary models. The extreme gradient boost model therefore had the best model for training the dataset, with a mean accuracy of 80.6% ( Figure 4).

Figure 4. Model accuracy metrics for all the models evaluated on the training data.

Prediction of modern contraceptive use

Using the remaining test data (predictions from unseen test data), the performance of the predictive models for predicting modern contraceptive use was compared using the mean accuracy and mean area under the curve (AUC) of the ML models in stratified tenfold cross-validation. After all the models were investigated in this study, extreme gradient boosting was found to have the greatest accuracy (81.97%, 95% CI{(79.06%, 82.7%)} as was the ROC area (76.63%), followed by logistic regression (80.52%, 95% CI {(78.6%, 82.3%)} and support vector machines (SVMs) (80.41%), 95% CI{(78.48%, 82.2%)}.

The XGBoost model had relatively low specificity (88.67%), which meant that it performed poorly in identifying predictors of modern contraceptive use in Ethiopia but had high sensitivity (66.6%), which meant that it was more accurate in identifying predictors of modern contraceptive use ( Table 5).

Table 5. Model accuracy metrics for all the models evaluated on the test data.

Confusion matrix	Logistic Regression		Random Forest		XG Boost		KNN		Naive Bayes		SVM
	Observed		Observed		Observed		Observed		Observed		Observed
Predicted	No MCP use	MCP use	No MCP use	MCP use	No MCP use	MCP use	No MCP use	MCP use	No MCP use	MCP use	No MCP use	MCP use
No MCP use	1179	181	1258	249	129	80	175	143	175	143	162	122
MCP use	163	243	84	175	233	494	187	431	187	431	200	452
Metrics	%		%		%		%		%		%
Accuracy	80.52		81.2		81.97		79.95		79.84		80.41
95% CI	(78.6, 82.3)		79.3, 83.0)		(79.06, 82.7)		(78.01, 81.8)		(77.89, 81.6)		(78.48, 82.2)
Sensitivity	57.31		41.27		66.60		56.84		44.34		64.86
Specificity	87.85		93.81		88.67		87.26		91.06		85.32
Positive predictive value	59.85		67.83		61.22		58.50		61.04		58.26
Negative predictive value	86.69		83.48		86.61		86.48		83.81		88.49
AUC	72.58		70.54		76.63		72.04		69.99		75.08

Visualization of the receiver operating characteristic (ROC) curve was performed. Among the six machine learning models employed in this study, the curve of the extreme gradient boost model had the highest percentage of AUC values, indicating that it is the best at classifying the use or nonuse of modern contraceptive methods among reproductive-age women in Ethiopia. Moreover, this best model represented an acceptable range of AUC values (76.63%) ( Figure 5).

Figure 5. Receiver operating characteristic (ROC) curve of the six models’ AUC percentages for comparison of model predictions on test data.

Discussion

In this study, a weighted sample of 8837 women of reproductive age was employed for the final analysis, which was limited to secondary data from the PMA Ethiopia 2019 Cross-sectional Household and Female Survey. The use of modern contraceptive methods was found to be extremely low (24%),³⁶ which is comparable to the findings of earlier studies carried out in Ethiopia. These earlier studies revealed that the use of modern contraceptive methods was 31.7% in rural Dembia District, northwestern Ethiopia⁵; 11.0% in the surrounding Peasant Association of Gondar Town⁵; 38.3% in Mojo Town, southern Ethiopia³⁷; and 67.4% in Hosanna.³⁸ The possible reason could be an increased expansion of government and private health institutions, including health posts, as well as the communication of information by health extension workers and various nongovernmental organizations (NGOs). The difference might also be due to differences in awareness of modern contraceptive methods.

To identify determinants of modern contraceptive use, each algorithm was trained on 80% of the total instances through random sampling, and its effectiveness was tested on 20% of the total instances through random sampling. Six widely used machine learning (ML) algorithms, logistic regression (LR), random forest (RF), K-nearest neighbors (KNN), extreme gradient boosting (XGBoost), naive Bayes (NB), and support vector machines (SVMs), were included in the study to predict determinants of modern contraceptive use among reproductive-age women in Ethiopia.

The performance of the predictive models in predicting determinants of modern contraceptive use was assessed using the remaining tested data (predictions from unseen test data) and compared against the mean accuracy and mean area under the curve (AUC) of the ML models in stratified tenfold cross-validation. Accordingly, extreme gradient boosting had the highest accuracy (81.97%), 95% CI (79.06%, 82.7%) and area under the ROC curve (76.63%). The performance of this model is much better than that of studies conducted on the prediction of contraceptive discontinuation among reproductive-age women in Ethiopia using the Ethiopian Demographic and Health Survey 2016 dataset. These studies used the random forest model as the best predictive model, with an accuracy of 68% and an ROC of 74% based on a tenfold cross-validation score on balanced training data³⁹; additionally, another study conducted in Ethiopia also found the random forest model to be the best machine learning model for predicting nutritional status for children under five years of age using EDHS data, with an accuracy and AUC of 68.2% and 0.76, respectively.⁴⁰ These results, however, are lower than those of an Indonesian study in which AdaBoost was identified as the most accurate model for predicting the duration of contraceptive use, for which the accuracy was 85.1%. The size of the dataset used to develop the model may be the cause of this mismatch. The Indonesian study employed 39,594 records, whereas this study used only 8837 records, allowing the model to train more effectively and make predictions with greater accuracy.

Specific characteristics related to the use of modern contraceptives in Ethiopia that can be used as intervention targets were compared, identified, and recognized with the aid of machine learning methods. The extreme gradient boosting (XGB) and support vector machine models have the highest prediction power among the constructed predictive models compared to other machine learning classifier models, such as the RF and KNN models. According to the extreme gradient boost feature importance results, the variables were age at first family planning use, marital status, partner/husband feelings about family planning and relationship with the head of household. In addition, age and health care providers spoke about FP methods at health facility visits; if I use FPs, side effects might disrupt my relationship, and household size is also an important predictor of modern contraceptive use. This study is roughly in line with previous findings.

Age at first family planning use (20-30 years and above) was the first significant characteristic of modern contraceptive use among reproductive-age women in Ethiopia. This finding was consistent with that of a study performed in Bangladesh, southern Ethiopia, Nigeria, and the Democratic Republic of the Congo, which revealed that as women’s age increased from 15 to 34, the likelihood that they would use contraceptives increased.^41–44 The most likely explanation is that in rural settings, this age range is when most women are involved in various activities to take care of their household’s requirements, leading them to wish to space their pregnancies. Therefore, they favor the use of contraceptive techniques. The other factor might be that people in this age group now have women’s forum associations to debate the topic, as well as greater experience sharing from colleagues and neighbors. As a result, their utilization rate could increase. These results, however, did not align with research carried out in Mojo town,³⁷ and a study performed in Kerman,⁴⁵ Iran, revealed that people who used modern contraceptives had a younger mean age than people who did not. Differences in socio-demographic characteristics and durations may also explain the differences.

The marital status of reproductive-age women in Ethiopia was one of the other most significant factors for predicting modern contraception use. This conclusion was consistent with the findings of studies conducted in Tanzania and Gondar, Ethiopia, which showed that married women were more likely to use modern contraceptives than unmarried (single, widowed, or divorced) women were.^4,5 The outcome highlights the significance of male involvement in reproductive health issues, such as fertility and contraception, as well as couples’ motivation through education. Counseling and FP education should encourage couples to share their fertility concerns with one another.

Couples’ desire to have a/another child was also the most important feature of modern contraceptive use in predicting reproductive-age women in Ethiopia. This finding is consistent with the findings of studies conducted on predictors of modern contraceptive method use among married women of reproductive age in Western Ethiopia and elsewhere, which showed that those respondents who did not express future desire for children were 2.6 times more likely to utilize modern contraceptives during the study period.^14,15 It was obvious that women who desired children were not ready to use contraceptives.

The most significant aspect of modern contraceptive use predictions among Ethiopian women of reproductive age was the partner/husband’s attitude toward family planning. These findings are consistent with research conducted in Gondar, Ethiopia.⁵ This may be attributable to the discussion that can lead to an appropriate decision regarding the selection and use of FP methods, and the fact that the discussion was present in rural areas suggested that there may be a high level of knowledge regarding FP methods.

The most significant aspect of modern contraceptive use prediction was the discussion of FP techniques by healthcare personnel during health facility visits. This result is in line with the findings of other studies performed elsewhere that emphasize the need to promote contraceptive use after delivery by utilizing the prenatal period as a window of opportunity. Effective contraception counseling included in comprehensive ANC not only boosts client satisfaction and prenatal care quality but may also lead to an increase in postpartum contraceptive use.^46,47 There is evidence suggesting that health care workers could be trained/retrained to provide more effective FP services through group health education sessions, the distribution of simple educational material to postpartum FPs, individualized counseling and the initiation of chosen contraceptive methods.^48,49

Conclusion and Recommendations

In this study, the utilization of modern contraceptive methods was found to be extremely low. Six widely accepted machine learning algorithms have been used to predict determinants of modern contraceptive use in Ethiopia. Different confusion matrices were used to compare the candidate supervised machine learning algorithms. Based on these results, extreme gradient boosting (XGB) was the best performing model. Age at first family planning use, marital status, partner/husband feelings about family planning and relationships with the head of household, women’s age, having discussions with healthcare providers about FP methods at health facility visits, and household size were important predictors of modern contraceptive use. The use of modern contraceptives is therefore expected to increase with effective contraceptive counseling during ANC/PNC follow-up on family planning use and increasing partner discussions on FP. Enhancing contraceptive counseling techniques concerning the age at which family planning is used for the first time and the engagement of men in FP should also be investigated. It was also necessary to consider enabling women to choose their methods through spousal discussion and providing health information to modify traditional attitudes around the number of children, which was seen as beneficial for the family.

Consent for publication

Not applicable.

Patient and public participation

Not applicable.

Author’s contributions

JBA, Conceived and designed the study; analysis, interpreted the result and wrote the paper. SDK, ADW and DNM made significant contributions to the work reported; contributed to the acquisition of data, contributed to all these areas; participated in drafting, revising or critically reviewing the article; and agreed to be accountable for all aspects of the work. All the authors read and approved the final manuscript.