Free online XLSX to ARFF converter

Weka and some data-mining applications require ARFF instead of Excel workbooks, so spreadsheet data must be converted before it can be analyzed. The conversion extracts a worksheet into one typed dataset; it does not preserve the workbook's layout or editing features.

What the XLSX format is

XLSX is the XML-based workbook format used by Microsoft Excel and supported by LibreOffice Calc, Apple Numbers, and Google Sheets when exporting to Excel format. It can contain multiple worksheets, formulas, dates, formatting, merged cells, charts, filters, and hidden rows.

XLSX files are common in business reports, surveys, laboratory records, data exports, and manually maintained datasets. Machine-learning software normally needs one rectangular range of records, so select one worksheet and remove titles, subtotals, notes, and other non-record rows before conversion.

What the ARFF format is

ARFF, or Attribute-Relation File Format, is a plain-text dataset format associated with Weka. Its header declares the relation and each attribute, followed by a @data section containing the records.

@relation customers

@attribute age numeric
@attribute region {north,south,west}
@attribute churn {yes,no}

@data
42,north,no
35,south,yes

Common attribute types include numeric, string, date, and finite nominal lists such as {yes,no}. ARFF uses ? for missing values. Because names, types, and category definitions are stored with the data, Weka can import the dataset without guessing all metadata from a separate spreadsheet. ARFF is also readable in a text editor and suitable for version control.

Convert with 101convert.com

1. Open the XLSX-to-ARFF converter on 101convert.com. The file can be converted online for free without installing software.
2. Upload the .xlsx workbook and select the worksheet containing the records if worksheet selection is available.
3. Start the conversion and download the resulting .arff file.
4. Open the file in Weka or a text editor. Check the relation name, attribute declarations, row count, first records, missing values, and inferred types before analysis.

Convert confidential workbooks only when the service's current file-handling terms are acceptable. For sensitive data, use a local workflow instead.

Convert locally with LibreOffice and Weka

A CSV intermediate file is useful when the worksheet needs visual cleanup before conversion.

1. Open the workbook in LibreOffice Calc and select the required worksheet.
2. Delete title rows, merged cells, explanatory notes, subtotals, blank trailing columns, and summary sections that are not observations. Ensure the first row contains unique attribute names and every data row has the same columns.
3. Choose File → Save As, select Text CSV (.csv), and configure UTF-8 encoding, comma as the field delimiter, and double quotation marks as the text delimiter.
4. Open Weka Explorer and choose Preprocess → Open file. Select the CSV file and inspect the attributes and instance count.
5. Correct incorrectly inferred types or remove unsuitable columns using Weka's preprocessing controls.
6. Use the Explorer's Save control and save the dataset as dataset.arff.

Some Weka releases or installations provide an Excel loader that can open XLSX directly. If Preprocess → Open file does not recognize the workbook, use the CSV route or install the required Weka loader and spreadsheet dependencies. Renaming .xlsx to .arff does not convert the file.

Convert programmatically with Python

Python is appropriate for repeatable conversions, worksheet selection, column renaming, and batch processing. Install an XLSX reader and an ARFF writer:

pip install pandas openpyxl liac-arff

import pandas as pd
import arff

df = pd.read_excel("input.xlsx", sheet_name="Data")
df.columns = [str(c).strip() for c in df.columns]

data = df.where(pd.notna(df), None)
attributes = []
for column in data.columns:
    values = data[column]
    if pd.api.types.is_numeric_dtype(values):
        kind = "numeric"
    else:
        data[column] = values.map(lambda value: None if value is None else str(value))
        distinct = sorted({value for value in data[column] if value is not None})
        kind = distinct if 0 < len(distinct) <= 50 else "string"
    attributes.append((str(column), kind))

with open("output.arff", "w", encoding="utf-8", newline="") as file:
    arff.dump({
        "relation": "dataset",
        "attributes": attributes,
        "data": data.values.tolist()
    }, file)

Replace sheet_name="Data" with a worksheet name or a zero-based index such as sheet_name=0. The example treats small, finite text columns as nominal and larger or free-text columns as strings; change the threshold and classifications to match the dataset. Review the generated ARFF because automatic type inference cannot determine whether an identifier is a useful feature, whether a numeric-looking code should remain text, or whether a category list is complete.

Quality and compatibility checks

• Formulas: ARFF stores values, not Excel formulas. Recalculate the workbook and save it before conversion if cached formula results may be stale.
• Worksheets: ARFF represents one relation. Convert each compatible worksheet to a separate file; do not combine sheets with different columns without a deliberate schema.
• Headers: Remove blank or duplicate headers and normalize excess spaces. Attribute names containing punctuation, spaces, commas, quotes, or line breaks may require quoting or escaping.
• Numeric data: Currency symbols, thousands separators, mixed units, and text such as unknown can cause a column to become text. Normalize these values before conversion.
• Missing values: Empty cells should become ?, not an empty string or zero, unless those values have the intended meaning.
• Nominal attributes: Every nonmissing value must appear in the finite declaration. Differences in spelling, capitalization, or trailing spaces create separate categories.
• Dates: Excel serial dates and locale-specific strings can be misread. Normalize them and declare an ARFF date pattern, for example @attribute created date 'yyyy-MM-dd'.
• Identifiers: Record IDs are often inferred as numeric or nominal attributes but usually should be excluded from predictive models because they do not describe the measured entity.
• Discarded content: Charts, formatting, comments, merged-cell layout, filters, workbook metadata, and formula expressions are not represented in ARFF. Preserve the original XLSX when those features are still needed.
• Encoding: Use UTF-8 and inspect non-ASCII names and values in Weka, especially when the data contains accented or non-Latin characters.