Use these indicators as the clearest decision-support results.
No indicators in this group for the current view.
Locality Filter All Mindanao records · 4500 of 4500 records
Review Setup
Select the service, stage, indicator, and predictor depth before reading the report.
Indicator-Level Classification Review
Classification is generated separately for each service indicator. This summary therefore focuses on
whether each indicator has enough evidence for interpretation, instead of presenting average scores
that may hide stronger and weaker indicator-level results.
The current dataset does not include direct access-barrier variables such as distance, waiting time, facility supply, staff interaction, or household service need. The portal should therefore improve the existing evidence first instead of inventing unavailable predictors.
Stage-specific modeling is applied through the saved stage response files and skip-pattern handling: Awareness uses valid awareness responses; Availment uses the availment stage; Satisfaction and Need for Action are interpreted only for respondents who reached the service-use stage.
The decision-tree pipeline now enriches the general profile and citizen-segment predictors with available service-context evidence when applicable: health background variables for Health Services, education background variables for Support to Education, and crime, disaster, corruption, and citizen-attitude evidence for Governance and Response. Feature selection remains part of the model search, while chi-square is kept in the citizen segment evidence area as profile-cluster support.
Use these as exploratory patterns and compare them with descriptive evidence.
No indicators in this group for the current view.
Do not overstate these indicators; they need more signal before stronger prediction claims.
- F1_FSSi - ChildandYouthWelfareProgram
- F1_FSSiv - Older Persons / Senior Citizens Program
- F1_FSSiii - PersonswithDisabilities (PWD)Welfare Program
- F1_FSSv - Family and Community Welfare Program
Recommended model comparison: keep Decision Tree as the official explanation model, then add Random Forest as a performance benchmark and Logistic Regression as a simple baseline. If another model improves ROC AUC or recall, report it as supporting evidence while keeping the decision tree for readable rules.
Model Insight
For Social Welfare Services - Awareness, the summary combines 6 indicator-specific decision-tree model(s). Mean F1 is 66.2% and mean ROC AUC is 51.6%, so the result should be read as an overall tendency across indicators. The indicator-level rows remain the main evidence for decision-support use.
Use the mean scores for the general story; use the indicator rows for the actual evidence.
Stage eligibility is handled before modeling: Awareness uses all valid Yes/No responses; Availment uses Awareness = Yes; Satisfaction and Need for Action use Awareness = Yes and Availment = Yes. Skip-pattern values such as 95-99 are excluded from the target class.
- Mean precision is 75.1%, indicating strong when an indicator model predicts the positive service outcome.
- Mean recall is 61.7%, meaning some indicator models may still miss actual positive cases, especially when recall is lower than precision.
- Mean F1 score is 66.2%, which balances precision and recall across the indicator models.
- Mean ROC AUC is 51.6%, showing how well the indicator models separate outcome groups across thresholds on average.
- Top predictors currently include MCA Dim2, Cluster 0, A5, MCA DimMagnitude, Cluster 1. These variables are useful signals for explaining which respondent profiles are linked to the selected service outcome.
- Interpretation: the model is more reliable when it says a respondent group is positive, but it may miss some groups that also belong to that outcome. Use it to prioritize follow-up, not to exclude citizens from attention.
Top Predictors
These are the strongest variables used by the saved decision-tree outputs for the selected service area and stage.
-
ChildandYouthWelfareProgram
MCA Dim2 -
ChildandYouthWelfareProgram
Cluster 0 -
ChildandYouthWelfareProgram
A5 -
Women’s Welfare Program
MCA DimMagnitude -
Women’s Welfare Program
MCA Dim2 -
Women’s Welfare Program
Cluster 1 -
PersonswithDisabilities (PWD)Welfare Program
A5 -
PersonswithDisabilities (PWD)Welfare Program
A3.1
Predictor Importance Chart
Show the ranked feature importance chart for the selected service-stage model.
Optional detail
Predictor Importance Chart
Show the ranked feature importance chart for the selected service-stage model.
Top Predictors
Recurring Predictors Across Decision Tree Models
This review counts which predictors repeatedly appear with positive importance across the saved indicator-level Decision Tree models. It helps identify citizen characteristics that consistently support interpretation across service-stage outcomes.
A recurring predictor is decision-support evidence, not proof of causation. Use it together with the indicator-level metrics, decision-tree example, clustering results, and service delivery review.
| Predictor | Predictor Group | Appeared in Models | Service Areas | Service Stages | Mean Importance | Plain Interpretation |
|---|---|---|---|---|---|---|
| MCA Dim2 | MCA profile dimension | 124 | 7 | 4 | 0.2751 | This is a combined profile signal from MCA. It should be read as a summary of related respondent characteristics, not as one survey question. |
| MCA Dim1 | MCA profile dimension | 113 | 7 | 4 | 0.2629 | This is a combined profile signal from MCA. It should be read as a summary of related respondent characteristics, not as one survey question. |
| MCA DimMagnitude | MCA profile dimension | 101 | 7 | 4 | 0.2349 | This is a combined profile signal from MCA. It should be read as a summary of related respondent characteristics, not as one survey question. |
| B4 | Housing and living conditions | 92 | 7 | 4 | 0.1659 | Household conditions or information access may be linked with how citizens know about and use LGU services. |
| B6 | Housing and living conditions | 92 | 7 | 4 | 0.155 | Household conditions or information access may be linked with how citizens know about and use LGU services. |
| A5 | Socio-demographic profile | 87 | 7 | 4 | 0.1811 | This respondent profile characteristic may help explain differences in awareness, service use, satisfaction, or need for action. |
| A3.1 | Socio-demographic profile | 75 | 7 | 4 | 0.151 | This respondent profile characteristic may help explain differences in awareness, service use, satisfaction, or need for action. |
| A1 | Socio-demographic profile | 59 | 7 | 4 | 0.126 | This respondent profile characteristic may help explain differences in awareness, service use, satisfaction, or need for action. |
| A7 | Socio-demographic profile | 56 | 7 | 4 | 0.1261 | This respondent profile characteristic may help explain differences in awareness, service use, satisfaction, or need for action. |
| B2 | Housing and living conditions | 53 | 7 | 4 | 0.1056 | Household conditions or information access may be linked with how citizens know about and use LGU services. |
| B1 | Housing and living conditions | 50 | 7 | 4 | 0.1258 | Household conditions or information access may be linked with how citizens know about and use LGU services. |
| B3 | Housing and living conditions | 44 | 7 | 4 | 0.127 | Household conditions or information access may be linked with how citizens know about and use LGU services. |
| A4 | Socio-demographic profile | 42 | 7 | 4 | 0.1198 | This respondent profile characteristic may help explain differences in awareness, service use, satisfaction, or need for action. |
| A2 | Socio-demographic profile | 30 | 7 | 4 | 0.0739 | This respondent profile characteristic may help explain differences in awareness, service use, satisfaction, or need for action. |
| A8 | Socio-demographic profile | 28 | 6 | 4 | 0.0957 | This respondent profile characteristic may help explain differences in awareness, service use, satisfaction, or need for action. |
Decision Tree Interpretation
Use this view to understand the main decision logic before opening the technical tree diagram.
How to read this result
Interpretation: among respondents with A8 <= 3.500, especially those in the associated MCA profile range, especially those in the associated MCA profile range, the model tends to predict awareness.
Some respondents with A8 > 3.500, especially those in the associated MCA profile range, especially those in the associated MCA profile range are predicted Not Aware.
Key technical terms
- First split The first condition is the strongest separator used by the simplified tree.
- Samples The number of eligible respondent records that reached a node.
- Value The distribution of respondents across the two possible outcome classes.
- Class The predicted outcome for that branch, such as aware or not aware.
Predictors shown in the simplified tree
MCA Dim2
MCACluster
A5
A8
Technical Tree Diagram Optional detail
Evidence Tables
Show readiness review, indicator metrics, rule reference, and diagnostics.
Optional detail
Evidence Tables
Show readiness review, indicator metrics, rule reference, and diagnostics.
Model Readiness Review
This table explains whether each indicator has enough signal for decision-support use and why some results should remain exploratory.
| Indicator | Signal Quality (F1 / ROC AUC / Recall / Baseline) | Meaning | Majority Response (Baseline Class) | Eligible Records | Baseline Accuracy (Majority Guess) | Model Accuracy | Gain (Model - Baseline) | Possible Cause | Suggested Data Improvement |
|---|---|---|---|---|---|---|---|---|---|
| F1_FSSi ChildandYouthWelfareProgram |
Needs more predictors | The model has difficulty separating respondents based on ROC AUC, recall, or F1. | Yes / Positive (88.6%) | 4500 | 88.6% | 84.2% | -4.4 pts | The result may be close to the majority-class baseline, meaning the model adds limited separation beyond the most common response. | Add or review awareness-source, barangay information channel, distance, and program availability variables for Social Welfare Services. |
| F1_FSSiv Older Persons / Senior Citizens Program |
Needs more predictors | The model has difficulty separating respondents based on ROC AUC, recall, or F1. | Yes / Positive (94.8%) | 4500 | 94.8% | 62.8% | -32.0 pts | The result may be close to the majority-class baseline, meaning the model adds limited separation beyond the most common response. | Add or review awareness-source, barangay information channel, distance, and program availability variables for Social Welfare Services. |
| F1_FSSiii PersonswithDisabilities (PWD)Welfare Program |
Limited signal | The model can be reviewed as an exploratory clue, but F1, ROC AUC, or baseline gain is not strong enough for confident prediction. | Yes / Positive (66.8%) | 4500 | 66.8% | 57.5% | -9.4 pts | The result may be close to the majority-class baseline, meaning the model adds limited separation beyond the most common response. | Add or review awareness-source, barangay information channel, distance, and program availability variables for Social Welfare Services. |
| F1_FSSv Family and Community Welfare Program |
Needs more predictors | The model has difficulty separating respondents based on ROC AUC, recall, or F1. | Yes / Positive (62.0%) | 4500 | 62.0% | 53.0% | -9.0 pts | The result may be close to the majority-class baseline, meaning the model adds limited separation beyond the most common response. | Add or review awareness-source, barangay information channel, distance, and program availability variables for Social Welfare Services. |
| F1_FSSvi Programs for Internally Displaced Persons |
Needs more predictors | The model has difficulty separating respondents based on ROC AUC, recall, or F1. | Yes / Positive (51.2%) | 3900 | 51.2% | 50.9% | -0.3 pts | The result may be close to the majority-class baseline, meaning the model adds limited separation beyond the most common response. | Add or review awareness-source, barangay information channel, distance, and program availability variables for Social Welfare Services. |
| F1_FSSii Women’s Welfare Program |
Needs more predictors | The model has difficulty separating respondents based on ROC AUC, recall, or F1. | Yes / Positive (80.7%) | 4349 | 80.7% | 39.1% | -41.6 pts | The result may be close to the majority-class baseline, meaning the model adds limited separation beyond the most common response. | Add or review awareness-source, barangay information channel, distance, and program availability variables for Social Welfare Services. |
Indicator-Level Metrics
How the mean scores are formed
The AVG values summarize 6 indicator model(s) for Social Welfare Services - Awareness. They describe the overall pattern across indicators, not the result of one specific indicator.
Average F1 is 66.2%, which gives the most balanced quick reading because it considers both correct positive predictions and missed positive cases.
Average ROC AUC is 51.6%, so the model should be read as decision-support evidence. Values closer to 50% mean the predictors have limited ability to separate the outcome groups.
Precision is high at 75.1%, but recall is lower at 61.7%. Positive predictions may be reliable, but the model may still miss some actual positive cases.
| Target | Indicator | Accuracy | Precision | Recall | F1 | ROC AUC | CV F1 | Best CV F1 | Selected Features | Base Predictors | Service Context | Total Predictors | Depth | Leaf | Criterion |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| F1_FSSi | ChildandYouthWelfareProgram | 84.2% | 88.6% | 94.2% | 91.3% | 51.5% | 76.0% | 86.8% | 8 | 17 | 0 | 17 | 3 | 10 | gini |
| F1_FSSiv | Older Persons / Senior Citizens Program | 62.8% | 94.5% | 64.3% | 76.6% | 49.2% | 78.2% | 79.1% | 8 | 17 | 0 | 17 | 6 | 35 | entropy |
| F1_FSSiii | PersonswithDisabilities (PWD)Welfare Program | 57.5% | 69.6% | 64.6% | 67.0% | 54.4% | 59.7% | 68.7% | 16 | 17 | 0 | 17 | 3 | 10 | gini |
| F1_FSSv | Family and Community Welfare Program | 53.0% | 66.0% | 53.0% | 58.8% | 53.6% | 68.9% | 67.9% | all | 17 | 0 | 17 | 3 | 35 | gini |
| F1_FSSvi | Programs for Internally Displaced Persons | 50.9% | 53.5% | 60.9% | 57.0% | 50.7% | 42.8% | 63.3% | 8 | 17 | 0 | 17 | 3 | 35 | entropy |
| F1_FSSii | Women’s Welfare Program | 39.1% | 78.5% | 33.0% | 46.5% | 50.0% | 74.6% | 77.0% | 16 | 17 | 0 | 17 | 3 | 35 | entropy |
Decision Rule Reference
Some tree rules use coded profile values. Read the tree from top to bottom, then use this reference to translate the rule into respondent-friendly meaning.
| Code | Meaning | Code Values / Variable Type | How to Read the Split |
|---|---|---|---|
| A8 | Place of Work | 1=W/in the barangay; 2=W/in the municipality; 3=W/in the province; 4=W/in the region; 5=W/in the country; 95=Not Applicable | <= 3.5 means W/in the barangay, W/in the municipality, W/in the province; > 3.5 means W/in the region, W/in the country, Not Applicable. |
| Dim2 | MCA profile dimension 2 | A combined respondent-profile score from MCA. It is not a single survey question. | Dim2 <= -0.301 follows one side of the profile map; values above -0.301 follow the other side. |
Gini Split Diagnostics
| Node | Type | Rule | Gini | Samples | Not Aware | Aware | Prediction |
|---|---|---|---|---|---|---|---|
| 0 | Split | A8 <= 3.500 | 0.5 | 4500 | 0.5 | 0.5 | Aware |
| 1 | Split | Dim2 <= -0.301 | 0.4945 | 1884 | 0.4 | 0.6 | Aware |
| 2 | Split | Dim2 <= -0.471 | 0.4906 | 1705 | 0.4 | 0.6 | Aware |
| 3 | Leaf | Prediction | 0.4938 | 1528 | 0.4 | 0.6 | Aware |
| 4 | Leaf | Prediction | 0.4158 | 177 | 0.3 | 0.7 | Aware |
| 5 | Split | Dim2 <= -0.201 | 0.4901 | 179 | 0.6 | 0.4 | Not Aware |
| 6 | Leaf | Prediction | 0.4399 | 67 | 0.7 | 0.3 | Not Aware |
| 7 | Leaf | Prediction | 0.4995 | 112 | 0.5 | 0.5 | Aware |
| 8 | Split | Dim2 <= 0.343 | 0.4978 | 2616 | 0.5 | 0.5 | Not Aware |
| 9 | Split | Dim2 <= 0.301 | 0.4824 | 767 | 0.6 | 0.4 | Not Aware |
| 10 | Leaf | Prediction | 0.488 | 738 | 0.6 | 0.4 | Not Aware |
| 11 | Leaf | Prediction | 0.2865 | 29 | 0.8 | 0.2 | Not Aware |
| 12 | Split | Dim2 <= 1.032 | 0.5 | 1849 | 0.5 | 0.5 | Not Aware |
| 13 | Leaf | Prediction | 0.496 | 1158 | 0.5 | 0.5 | Aware |
| 14 | Leaf | Prediction | 0.4896 | 691 | 0.6 | 0.4 | Not Aware |