Demand Forecasting Dashboard

Produk:

Window (n):

Alpha (α):

Delta (δ):

SMA Simple Moving Average (n periode)

F_t = (A_t-1 + A_t-2 + ... + A_t-n) / n

Keterangan:
• F_t = Forecast untuk periode t
• A_t-i = Actual sales i periode sebelumnya
• n = Jumlah periode (window size)

Contoh (n=6, Bulan 7):
F₇ = (A₁ + A₂ + A₃ + A₄ + A₅ + A₆) / 6

WMA Weighted Moving Average

F_t = (1×A_t-n + 2×A_t-n+1 + ... + n×A_t-1) / Σweights

Keterangan:
• Bobot: 1, 2, 3, ..., n (periode terbaru = n)
• Total bobot = n(n+1)/2
• Data terbaru memiliki pengaruh lebih besar

ES Exponential Smoothing [18.3]

F_t = F_t-1 + α(A_t-1 - F_t-1)

Keterangan:
• F_t = Forecast periode t
• F_t-1 = Forecast periode sebelumnya
• A_t-1 = Actual demand periode sebelumnya
• α = Smoothing constant (0 < α < 1)

Interpretasi:
Forecast baru = Forecast lama + α × (Error forecast lama)

HOLT Exponential Smoothing with Trend [18.4-18.6]

[18.4] F_t = FIT_t-1 + α(A_t-1 - FIT_t-1)

[18.5] T_t = T_t-1 + δ(F_t - FIT_t-1)

[18.6] FIT_t = F_t + T_t

Keterangan:
• F_t = Smoothed forecast (tanpa trend)
• T_t = Smoothed trend
• FIT_t = Forecast Including Trend (forecast final)
• α = Smoothing constant untuk level
• δ = Smoothing constant untuk trend

LR Linear Regression

Y = a + bX

b = (nΣXY - ΣXΣY) / (nΣX² - (ΣX)²)

a = (ΣY - bΣX) / n

Keterangan:
• Y = Nilai forecast (demand)
• X = Periode waktu (1, 2, 3, ...)
• a = Intercept (nilai Y ketika X = 0)
• b = Slope (kemiringan garis, trend per periode)
• n = Jumlah data

Interpretasi:
Jika b > 0 → Trend naik | Jika b < 0 → Trend turun

Rumus Error Metrics

Error: e_t = A_t - F_t

MAD: Σ|e_t| / n

MSE: Σ(e_t)² / n

MAPE: (Σ|e_t/A_t|) × 100 / n

Tracking Signal: TS = RSFE / MAD

Interpretasi Tracking Signal:
• |TS| ≤ 4 → Forecast dalam kontrol ✓
• |TS| > 4 → Forecast bias, perlu dikoreksi ✗

Data Aktual Penjualan (12 Bulan)

Grafik Perbandingan Forecast

Analisis Error

Error per Bulan

Tracking Signal

Perhitungan Forecast & Error

Kesimpulan