Bungalow Valuation in Kemaman

No black box. Here’s the full pipeline, step by step, with the actual formulas and thresholds used in our code.

1. 1
   Pull comparable transactions

   We query JPPH subsale records matching your selection:

   • Exact match on state / district / mukim / area (or a list of areas when scoped by township/postcode).
   • Exact match on property type (e.g. “condo” maps to Condominium/Apartment; “terrace” covers 1/1½ & 2/2½ storey terraced + low-cost).
   • Last 3 years of transactions only (date ≥ today − 3 years).
   • Progressive size tolerance: try ±30% of your sqft first, widen to ±50%, then drop the size filter — whichever yields at least 3 comparables.
   • Capped at 50 most recent transactions. Minimum 3 required; 1–2 falls back to “rough estimate” mode.
2. 2
   Strip outliers (batch cap + IQR on PSF)

   Developer bulk transfers and anomalous sales get removed in two phases:

   • Batch cap: if the same date + area has more than 5 transactions, we keep only the 5 highest-priced ones (bulk SPAs from developers often cluster at the same date+area and drag the median down).
   • IQR filter on PSF: compute price-per-sqft for each, sort, then
     Q1 = 25th percentile, Q3 = 75th percentile
     IQR = Q3 − Q1
     keep if Q1 − 1.5×IQR ≤ PSF ≤ Q3 + 1.5×IQR

   You can disable this with the “Include outliers” toggle — the scatter chart in the result shows which points were kept (blue) vs removed (red).

3. 3
   Weight each transaction by recency (exponential decay)

   Newer sales count more than older ones. Each comparable gets a weight:

   weight = e^{−0.05 × age_months}

   Age	1 mo	6 mo	12 mo	24 mo	36 mo
   weight	0.95	0.74	0.55	0.30	0.17

   Can be disabled via the “Recency weighting” toggle — then it’s a plain unweighted median.

4. 4
   Compute the weighted median price-per-sqft

   We use median, never average (averages get wrecked by a single luxury sale). Weighted median = sort PSF values by magnitude, walk cumulative weight, pick the value where cumulative weight crosses 50% of total weight.

   PSF_i = price_i ÷ effective_area_sqft_i
   weighted_median_PSF = PSF_k where ∑_i=1..k weight_i ≥ ½ ∑ weight_i
5. 5
   Multiply by your built-up size
   base_price = weighted_median_PSF × your_size_sqft

   Why PSF-based? Location and property type are already baked into the comparable filter, so multiplying PSF by your size is the cleanest way to scale to your specific unit without double-counting.

6. 6
   Apply four adjustments (multiplicatively)
   a. Tenure

   Freehold commands ~10% premium over leasehold. But some of that is already baked into the comparables, so we only apply the gap:

   freehold_ratio = freehold_comps ÷ total_comps
   user_premium = 0.10 if freehold else 0
   implicit_premium = freehold_ratio × 0.10
   tenure_mult = 1 + (user_premium − implicit_premium)

   b. Floor level (high-rise only)

   Split comparables with floor data into lower/upper halves by floor, measure the PSF gap, extrapolate to your floor vs the median floor. Capped at ±2%/floor. Fallback of 0.5% per floor above/below median when fewer than 10 comps have floor data.

   premium_per_floor = (upper_PSF ÷ lower_PSF − 1) ÷ floor_span
   floor_mult = 1 + (your_floor − median_floor) × premium_per_floor

   c. Market sentiment (±15% cap)

   Compare avg price of the 10 most recent comparables vs the next 10 older ones.

   momentum = (recent10_avg − older10_avg) ÷ older10_avg
   sentiment = clamp(0.5 + momentum, 0.1, 0.9)
   sentiment_adj = (sentiment − 0.5) × 0.15 × price

   d. Market trend (±10% cap)

   Linear regression (OLS) on price vs ordinal index of comparables, normalised to average price.

   slope = (n∑xy − ∑x∑y) ÷ (n∑x² − (∑x)²)
   trend = slope ÷ avg_price
   trend_adj = trend × 0.10 × price
   estimate = base_price × tenure_mult × floor_mult + sentiment_adj + trend_adj
7. 7
   Build the price range

   We use 15% of the standard deviation of comparable prices as a one-sigma proxy:

   σ = stdev(comparable_prices)
   margin = 0.15 × σ
   low = max(0, estimate − margin)  /  high = estimate + margin
8. 8
   Score the confidence (0.10–0.95)

   Starts at 0.5, then adds/subtracts based on four signals:

   Signal	Contribution
   Comparable count	+ min(0.4, count ÷ 50 × 0.4)
   Share < 6 months old	+ (recent ÷ total) × 0.2
   Price consistency (low CV)	+ max(0, 0.2 − σ÷μ)
   Sane location premium (0.5–2.0×)	+ 0.1

   confidence = clamp(sum, 0.10, 0.95)

   Labels: ≥0.8 = High · ≥0.6 = Medium · ≥0.4 = Moderate · <0.4 = Low.

Same estimate, different playbook depending on why you’re here. Pick the one that fits.