About the Project (Detailed Scientific Documentation)

We compute startYear and then sum days_above_40 from (startYear + 1) to maintenanceYear:

startYear = maintenanceYear - yearsSpan

HTS = Σ_{y = startYear + 1 إلى maintenanceYear}  days_above_40(y)
        

This matches the Laravel logic: whereBetween(year, [startYear+1, maintenanceYear]) then sum(days_above_40).

Because research reports that when ambient temperature exceeds 40°C, pavement surface temperature can reach ~65–75°C, significantly increasing the risk of rutting, deformation, and reduced load-bearing capacity. Therefore, counting “days above 40°C” becomes a meaningful simplified proxy for long-term thermal exposure—especially for Iraq/Karbala climate.

• • Source mentioning 65–75°C surface range when ambient >40°C and rutting risk. — MDPI (2025)
• • evalfis and FIS evaluation concept (HTS later becomes an ANFIS input). — MathWorks evalfis

Research note: HTS here is a designed proxy rather than a universal standard name. The goal is to represent climate severity (heat/time) numerically for the model.

We use log10(x + 1) to compress large ranges, reduce skewness, and improve numerical stability before normalization and evaluation—common when inputs span wide numeric ranges.

x_log = log10(x + 1)
        

Implementation note: log10_safe protects against unexpected non-positive values to avoid runtime failures.

After log transform, we apply the same MATLAB normalization using PS_in. In MATLAB, mapminmax can return process settings (PS) which can be reused to transform new data identically—ensuring the PHP implementation matches MATLAB exactly.

• • PS_in.gain / xoffset / ymin / yrange are used to map each feature to the target range.
• • You stored these values in config(anfis.ps_in.*) for reproducibility.

Here we evaluate the Sugeno FIS: compute membership degrees per rule, compute rule firing strengths using an AND operator (e.g., product), compute consequent outputs (often linear), then aggregate in a Sugeno manner to produce yNorm—conceptually matching MATLAB evalfis.

• • Sugeno FIS documentation. — MathWorks sugfis
• • evalfis documentation (FIS evaluation) as a reference concept. — MathWorks evalfis

After yNorm, we restore the original output scale using PS_out (reverse mapminmax) to get outLog. Then apply inverse log to obtain final cost:

outLog = reverse_mapminmax(yNorm, PS_out)

cost = (10 ^ outLog) - 1
        

Thus you implemented: Log → mapminmax → Sugeno eval → reverse mapminmax → inverse log, mirroring MATLAB.

Research value: you did not retrain the model on the web; you converted a trained MATLAB model into PHP while preserving the same transforms and parameters (fis + PS settings). This enables an accurate MATLAB vs PHP comparison.

A = TotalCost / Area   (IQD per m²)
        

The service first validates Area > 0, computes A, then matches it against configured ranges in config(cost_gauge.ranges).

• • If A is within [min, max), select that range.
• • If A is below the first range → use the first range.
• • If A exceeds the last range → use the last range.
• • Result returns label_key + label + range_used + full ranges for gauge rendering.

Cost per unit area is commonly used in cost estimation as a parametric normalization metric for comparison. — RAIC CHOP (Cost Planning)

1) Base cost:
   C0 = f(x)

2) For each feature i:
   - If continuous: test (xi - 10%) and (xi + 10%)
   - If categorical: test all allowed values (except current)

3) Max absolute impact:
   Δi = max | f(x_test) - C0 |

4) Impact percentage:
   Impact%i = (Δi / C0) * 100

5) Sort descending -> bar chart + top influencers
        

Finally, results are sorted by impact_percent (DESC). The top_n influencers are used to support the ExplanationService.

• • OAT concept and “perturbations typically occur one at a time”. — Razavi et al., 2021 (ScienceDirect)
• • Statement: “each input feature is perturbed one-at-a-time”. — Naik et al., 2021 (Springer)
• • Model-agnostic XAI methods based on perturbation and observing prediction changes. — Wikle et al., 2022 (PMC)

• • Gauge: classification + A index (cost/m²).
• • Sensitivity: top_influencers + debug_payload (direction, delta, ...).
• • inputsAssoc: current input values to describe the current state.

- min_influence_pct: أقل نسبة تأثير لقبول السبب
- max_reasons: الحد الأعلى لعدد الأسباب
- نمر على top_influencers:
    إذا pct >= minPct -> نولد سبب (AR + EN)
- إذا ماكو أسباب قوية -> fallback sentence
        

The service also determines direction (increase/decrease/neutral) using delta = cost_after_change - base_cost, then uses that to craft an engineering sentence explaining whether the current value is cheaper or costlier than alternatives within the tested range.

• • Consistent formal tone (good for academic reporting).
• • Auditable: each sentence maps to specific data (debug payload).
• • Maintainable: changing templates/rules is easier than training an NLG model.
• • Supports bilingual output easily (AR/EN).

These references are not your specific conversion, but official examples of MATLAB web app hosting/concepts that can support the “conversion chapter” context. — MathWorks Web App Server