Automated Time-Series Preprocessing in R • atspR

Automated Time-Series Preprocessing in R

Overview

atspR is an R package that provides an automated, modular preprocessing pipeline for time-series data covering the full journey from raw sensor data to cross-validated, model-ready train and test sets.

The package is designed for students and domain practitioners (agronomists, environmental scientists, irrigation engineers) who work primarily in R and need a reliable, leakage-free preprocessing workflow without having to manage multiple packages or worry about the correct sequencing of steps.

Why atspR?

Problem	How atspR solves it
Scalers applied before splitting → data leakage	Scaler fitted on TRAIN only, applied to both sets
Random split invalidates time-series models	Temporal split — train always precedes test chronologically
Irregular sensor timestamps break time-series assumptions	`fill_time_gaps()` snaps data onto a regular time grid, flagging missing points as `NA`
Silent NA removal with no record	Full report of what was dropped or imputed, and why
Multi-package workflow, easy to sequence incorrectly	Single function `ts_preprocess()` runs all 8 steps in order
One-size-fits-all model choice	`cross_validate()` supports multiple model types (Linear Regression (lm) / Generalized Additive Model (gam) / Random Forest (rf)/ Decision Tree (dt))
Error messages with no guidance	Every automated decision explained in plain-language output

Pipeline

Raw data
   │
   ├─ (Stage 1) combine_datetime()      Merge date + time columns → POSIXct (optional)
   ├─ (Stage 2) fill_time_gaps()        Insert placeholder rows for missing timestamps
   │
   └─ ts_preprocess()                   ────────────────── 8 steps ──────────────────────────────
   │    (Stage 3) standardize_na()      Convert sentinel values to NA
   │    (Stage 4) coerce_numeric()      Parse character columns to numeric
   │    (Stage 5) split_data()          Temporal train/test split
   │    (Stage 6) missing_analysis()    Summarise NAs; decide DROP or IMPUTE
   │    (Stage 7) handle_missing()      Drop rows or interpolate (linear/KNN)
   │    (Stage 8) visualize_data()      Scatter plots per variable vs time
   │    (Stage 9) scale_data()          MinMax / Z-score / Robust (auto)
   │    (Stage 10) cross_validate()     Walk-forward k-fold CV (optional) — choose one :
   |                                    Linear Regression (lm)/ Generalized Additive Model (gam)/
   │                                    Random Forest (rf)/ Decision Tree (dt)
   │
   └─ ts_export()                       Write all outputs to CSV

Installation

# Install from GitHub
# install.packages("devtools")
devtools::install_github("EasternTechFusion/atspR")

Quick Start

Note: Column names must not contain parentheses ( ) or the package will fail to parse them. For example, rename WaterLevel() to WaterLevel before passing the data to atspR functions.

Univariate

Sample data

DateTime	WaterLevel
2024-01-01 08:00	3.31
2024-01-01 08:10	3.37
2024-01-01 08:20	3.46
2024-01-01 08:33	3.53
2024-01-01 08:42	3.56
2024-01-01 08:59	3.58
2024-01-01 09:10	3.60
2024-01-01 09:24	3.62

Raw sensor readings arrive at irregular intervals (roughly every 10 minutes) — fill_time_gaps() snaps them onto a regular 10-minute grid, inserting NA for any grid point with no matching reading.

This dataset has only one variable (WaterLevel), so ts_preprocess() runs in univariate mode — no exogenous features are used.

library(atspR)

df$datetime <- as.POSIXct(df$datetime, 
                          format = "%Y-%m-%d %H:%M:%S")

gap <- fill_time_gaps(df,
                      time_col = "DateTime",
                      n        = 10,
                      unit     = "min")

result <- ts_preprocess(data          = gap$data,
                        train_ratio   = 0.8,
                        impute_method = "linear",
                        lags          = 24,
                        model_type    = "lm",
                        target_col    = "WaterLevel",
                        k_folds       = 5)

Multivariate

Case 1: Date + Time in separate columns

Sample data

Date	Time	Temp	Humid	Solar	WindSpeed	RainFall	VPD
2024-01-01	08:00	25.12	80.04	14.0591	0.0000	0.0000	0.2758
2024-01-01	09:00	NA	82.38	18.3274	0.5200	0.0000	0.3124
2024-01-01	11:00	26.26	85.03	22.7810	1.1400	0.0000	0.3892
2024-01-01	12:00	27.07	83.57	NA	1.3300	0.1200	0.4015

Hour 10:00 is missing entirely — fill_time_gaps() will insert it as a row of NA.

library(atspR)

df <- combine_datetime(my_data,
                       date_col  = "Date",
                       time_col  = "Time",
                       new_col   = "datetime",
                       time_type = "string")

gap <- fill_time_gaps(df, 
                      time_col = "datetime", 
                      n        = 1, 
                      unit     = "hour")

result <- ts_preprocess(data         = gap$data,
                        train_ratio  = 0.8,
                        target_col   = "VPD",
                        k_folds      = 5)

Case 2: Datetime in one column

Sample data

DateTime	Temp	Humid	Solar	WindSpeed	RainFall	VPD
2024-01-01 08:00	25.12	80.04	14.0591	0.0000	0.0000	0.2758
2024-01-01 09:00	NA	82.38	18.3274	0.5200	0.0000	0.3124
2024-01-01 11:00	26.26	85.03	22.7810	1.1400	0.0000	0.3892
2024-01-01 12:00	27.07	83.57	NA	1.3300	0.1200	0.4015

Hour 10:00 is missing entirely — fill_time_gaps() will insert it as a row of NA.

library(atspR)

df$datetime <- as.POSIXct(df$datetime, 
                          format = "%Y-%m-%d %H:%M:%S")
                          
gap <- fill_time_gaps(df, 
                      time_col = "datetime", 
                      n = 1, 
                      unit = "hour")
                      
result <- ts_preprocess(data         = gap$data,
                        train_ratio  = 0.8,
                        target_col   = "VPD",
                        k_folds      = 5)

Case 3: Date only (daily data)

Sample data

date	Temp	Humid	Solar	WindSpeed	RainFall	VPD
2024-01-01	28.38	75.04	14.0591	0.0000	0.0000	0.275835
2024-01-02	29.16	72.05	NA	0.5200	0.0000	0.312410
2024-01-04	27.89	78.03	22.7810	1.1400	2.4000	0.389200
2024-01-05	26.50	80.54	19.4320	0.8800	0.0000	0.301770

2024-01-03 is missing entirely — fill_time_gaps() will insert it as a row of NA.

library(atspR)

gap <- fill_time_gaps(df, 
                      time_col = "date", 
                      n = 1, 
                      unit = "day")
                      
result <- ts_preprocess(data         = gap$data,
                        train_ratio  = 0.8,
                        target_col   = "VPD",
                        k_folds      = 5)

Export Results

ts_export(result, dir = "C:\Users\user\", prefix = "data")

Cross-Validation

atspR uses walk-forward validation — the correct approach for time-series data. Each validation fold is always preceded only by past data, so no future information leaks into training.

Seed(20%)  Fold1  Fold2  Fold3  Fold4  Fold5
[─────────][─────][─────][─────][─────][─────]

Fold 1: train = seed              → val = fold1
Fold 2: train = seed + fold1      → val = fold2
Fold 3: train = seed + fold1+2    → val = fold3
...

Citation

Sueppong Mueanchamnong and Pattharaporn Thongnim (2026). atspR: Automated Time-Series Preprocessing in R.
R package version 1.1.0
https://github.com/EasternTechFusion/atspR

atspR

Overview

Why atspR?

Pipeline

Installation

Quick Start

Univariate

Multivariate

Case 1: Date + Time in separate columns

Case 2: Datetime in one column

Case 3: Date only (daily data)

Export Results

Cross-Validation

Citation

License